Improved target detection method of the constant false alarm rate type for a multi-channel detection system.

Abstract

An improved method for constant false alarm rate target detection in a multi-channel detection system. A computer-implemented method (100) for constant false alarm rate target detection (C) in a multi-channel detection system, characterized in that, considering a short observation period for any target to be considered stationary, for each sampling step (j) along a distance direction (D) of the multi-channel detection system, a detection test is performed on a sequence of hybrid signals along a recursive direction of the multi-channel detection system, the sequence covering the observation period and each hybrid signal (i) associating a sum channel signal (i) and a deviation signal (d) for each sampling step (i) along the recursive direction, the method generating target detection when the detection test is verified. Figure for abstract: Figure 2

Description

Title of the invention: Improved method for constant false alarm rate type target detection for a multi-channel detection system.

[0001] The invention relates to the field of target detection methods for multi-channel detection systems of the radar or sonar type.

[0002] A target detection method delivers a list of detections or "plots" at each instant. Downstream detection methods then correlate the plots from one instant to the next to open, update, or close leads. Each lead is finally analyzed to determine if it is a target of interest.

[0003] Target detection methods with constant false alarm rate -TFAC are known for multi-channel detection systems.

[0004] For example, in a two-way antenna, the physical antenna is "cut" into two sub-antennas, preferably of the same size, each sub-antenna being associated with electronics allowing the signal it receives to be digitized.

[0005] Target detection is then carried out using the complex signal from the sum channel, i.e., resulting from the summation of the complex digitized signals received by each of the two sub-antennas:

[0006]

[0007] More precisely, for a given radar distance, we consider a time sequence of N signals on the sum channel corresponding to the echoes collected during N successive recurrences: 100081 5^=(¾ .... s s ...,s w ] T , S^X

[0009] where i is an integer between 1 and N.

[0010] When the detection system uses different frequencies from one recurrence to the next, the TFAC detection processing relies on a detection test T1 comparing the sum of the powers of the signals on the sum channel at each instant to a threshold Sp:

[0011] T , .2 ç 1 i<ôi

[0012] When the same frequency is used from one recurrence to the next, the TFAC detection processing relies on a detection test T2 comparing, to a second threshold S2, the squared magnitude of the sum of the signals on the sum channel at each instant, up to a phase shift: 100131 T2:ie[ftN-l

[0014] This last equation is recognized as the expression of a Fourier transform.

[0015] When the detection test is verified, the detection process creates a "plot" (or detection).

[0016] This plot is characterized by a position of the target in bearing, in elevation and in distance relative to the detection system.

[0017] The TF AC detection processing carries out detection (or pre-detection in some cases) in "short time", that is to say on a time horizon of the order of tens to hundreds of milliseconds, corresponding to the usual transit time of the antenna lobe over the target in the case of a progressive scanning radar for example.

[0018] The position information attached to a plot is obtained taking into account the observation direction of the detection system at the time of the acquisition of the echoes which led to the verification of the detection test.

[0019] Since the main lobe of a detection system has a certain opening (generally defined at -3 dB), the position information of a plot has an accuracy substantially equal to half the opening of the main lobe of the detection system.

[0020] Moreover, and completely independently of the detection processing, the majority of known detection systems carry out a fine estimation of the direction of arrival of the wave on the detection system by a multi-channel deviation processing (called "monopulse radar" in English).

[0021] For the simple case of a two-channel system, a possible deviation processing consists of combining the signal on the sum channel (¾ = ^1+¾) and the signal on the difference channel (^ = according to the relation:

[0022] c _<s2, j.sA> Se~ N2

[0023] The expression in which the deviation signal SE is defined as the dot product of the signal on the sum channel by the signal on the phase-shifted difference channel of

[0024] It is then shown that:

[0025] sE= -tan^siniG)]

[0026] With e, the distance between the phase centers of the first and second sub-antennas; A, the wavelength of the carrier frequency of the received echo; and G, the angle of arrival of the echo evaluated with respect to the direction normal to the plane of the antenna in a plane defined by the direction normal to the plane of the antenna and the direction passing through the phase centers of the first and second sub-antennas.

[0027] For small arrival angles G, the arrival angle of the echo can then be estimated according to the following relation:

[0028] G=À-atan(-SF)

[0029] When the antenna plane is arranged horizontally, the arrival angle G is the bearing angle and when the antenna plane is arranged vertically, the angle G corresponds to the elevation angle.

[0030] In other words, with for example an antenna cut into four sub-antennas distributed along both a horizontal direction and a vertical direction, it is possible to simultaneously measure the bearing angle and the elevation angle of the direction of arrival of a signal.

[0031] Thus, the multi-channel detection system includes a module adapted to determine, for each echo of a succession of N echoes, an angle of arrival, Q..

[0032] This angular information is transmitted to a module called the pad dressing module, downstream of the module performing the TF AC detection processing and generating the pads. The pad dressing module is adapted to associate an arrival angle with each pad identified at the output of the TFAC module.

[0033] The cladding module thus supplements the position information of a stud with information on the angle of arrival. This allows for refining the angular position information of the stud as determined by the TFAC module.

[0034] Thus, according to the prior art, the deviation measurement is not taken into account in the TFAC detection processing.

[0035] However, the deviation measurement carries information that it would be desirable to use for target detection as such.

[0036] The present invention therefore aims to address this problem.

[0037] For this purpose, the invention relates to a computer-implemented method for constant false alarm rate (CFAR) target detection for a multi-channel detection system, characterized in that the method, considering an observation period short enough for any target to be considered stationary, implements, for each sampling step along a distance direction of the multi-channel detection system, a detection test on a sequence of hybrid signals along a recursive direction of the multi-channel detection system, the hybrid signal sequence covering the observation period and each hybrid signal associating a sum channel signal and a deviation signal for each sampling step along the recursive direction, the method generating a target detection when the detection test is verified.

[0038] According to particular embodiments, the process comprises one or more of the following characteristics, taken individually or in all technically possible combinations: - the detection test corresponds to an optimal test T ^G, the optimal test being written:

[0039] > c

[0040] with A a likelihood function of the time SpG sequence of hybrid instantaneous signals either under the assumption Ho of the presence of a target C or under the assumption of the absence of a target, and a hybrid detection threshold, the method leading to the generation of a detection, or plot, when the detection test is verified, the hybrid detection threshold being calculated to respect a constraint false alarm probability.

[0041] - the method comprises, in the case of a multi-channel scanning detection system, a filtering step allowing the calculation of a compressed arrival angle Gcomp from the deviation signal Q according to the relation: 100421 GcompAP = 2^,0^(7)^(^-0 + ^+1)

[0043] with h a time filter adapted for searching for a fixed target over a passage time of a main lobe of an antenna of the multi-way detection system over the target, N the number of hybrid signals in the sequence, i a sampling step index along the recurrence direction, j a sampling step index along the distance direction, and q an integer summation index.

[0044] - the method includes a post-integration step of the sum channel signal to obtain a compressed sum channel signal S^pj:

[0045] _ ji / -m2 U) ~ NIVJ ' I

[0046] with N the number of hybrid signals in the sequence, i a sampling step index along the recurrence direction, j a sampling step index along the distance direction, and q an integer summation index.

[0047] - the process includes a dimension reduction step allowing combination The compressed sum channel signal Syp^ and the compressed deviation signal Gcomp use a predefined metric to obtain a scalar value, and a step compares the scalar value to the hybrid detection threshold Spp

[0048] - the metric is defined by:

[0050] with ^ppp a covariance matrix,

[0051] - the method makes it possible to search for abnormally strong echoes and abnormally stable in angle of arrival over a short observation period, less than a second, and a short sampling step, less than ten milliseconds.

[0052] The invention also relates to a multi-channel detection system incorporating a computer programmed to execute the steps of the preceding process.

[0053] Preferably the system is of the radar type or of the sonar type.

[0054] The invention also relates to a computer program comprising program code instructions for executing the steps of the preceding process.

[0055] The invention and its advantages will be better understood upon reading the following detailed description of a particular embodiment, given solely by way of non-limiting example, this description being made with reference to the accompanying drawings in which: - [Fig.1] The [Fig.1] is a schematic representation of the operational situation in which a main lobe of a multi-channel radar system scans an area of ​​interest around a direction of interest; - [Fig.2] The [Fig.2] is a schematic block representation of a preferred embodiment of the process according to the invention; - [Fig.3] The [Fig.3] present different images obtained during the implementation of the process of [Fig.2].

[0056] The method according to the invention is based on the use of the estimated arrival angle Q, obtained by deviation measurement, within the TF AC detection processing itself, in addition to the signal from the sum channel.

[0057] The invention therefore relates to the native use, from the short-term target detection processing, of a sequence of estimated arrival angles { , obtained on a observation time At 1S, within a hybrid AC TF detection processing.

[0058] Generally, with the method according to the invention, target detection is performed on a hybrid signal S^q-

[0059] The hybrid signal comprises hybrid elementary signals for a sequence of N recurrences (i being an integer between 1 and N).

[0060] Each hybrid elementary signal S^Gi is two-dimensional: it associates the signal on the sum channel Sjj and the estimated arrival angle Q,.

[0061] We therefore have:

[0062] S^G = [ ^Gl • • • sZG,i • • • SSG,N ] =

[0063] In a preferred embodiment implementing a scanning radar, the sequence of N recurrences is obtained over an observation time interval At which corresponds to the scanning time of an azimuth of interest Az by the main lobe of the SS,1 ••• $2,N Gi ... G,... Gn Radar system antenna diagram. The temporal sampling step (or recurrence step) is typically 1 ms. The number N of signals in the sequence is typically 10. Thus, the observation period At is typically 10 ms.

[0064] As illustrated in Figure 1, the main lobe 10 is directed along a pointing direction D. It sweeps a horizontal plane defined by a coordinate system XY centered at point O where the radar antenna is located. If the direction X coincides, for example, with North, the main lobe 10 sweeps the XY plane with a sweep speed VBaB

[0065] The antenna diagram, limited to its main lobe for simplicity of description, has an angular width of -3dB, denoted 0dB'

[0066] A direction of interest A (for example, azimuth 90° in the example of the figure 1) is then covered by the main lobe 10 of the radar antenna during the observation period At.

[0067] We thus have the following set of extended equations:

[0068] Az (t) = Az0 + VBal.t

[0069] 0 <t<At AÜ<§ls

[0070] VBalAt^63dB

[0071] With Az(fy) the azimuth of the pointing direction D of the antenna at time t; A-^o, an azimuth of the pointing of the antenna at t = 0s; At, the sweep time over an angle of ^3dB'

[0072] In the case of the implementation of a fixed radar, the radar antenna observes in a fixed pointing direction, without scanning (VBaj = 0 F Bd ! s)-

[0073] We then have the following set of reduced equations:

[0074] Az(t)=Az0

[0075] 0 <t<At Aù^ls

[0076] By making, initially, the assumption Ho of the presence of a target C along the direction of interest A (more precisely of a distant target, of low speed and of sufficient signal-to-noise ratio - SNR), we can then consider the apparent azimuth of the target, Azc, as constant during the observation time Af.

[0077] The estimated bearing angle sequence J p 1 can then be modeled as follows:

[0078] g. = Azc-Az0- VBa?-^.At+ Ui

[0079] With: Azc, the azimuth of the target; and, uï, a noise on the measurement of the angle of arrival.

[0080] The measurement noise can be described by a normal N distribution of value mean zero and standard deviation (j^:

[0081] urN (0, ¢7¾ (SNRC))

[0082] The standard deviation (j^ depends on the signal-to-noise ratio of the target SNR(j): it is reduced if the SNR of the target is high, and increases when the SNR of the target decreases and gradually gets lost in the background noise.

[0083] By subsequently assuming the absence of a target along the direction of interest A, noises, such as background noise from sea surface echoes (also called "clutter") or thermal noise in the antenna's receiving circuits, generate a sequence of estimated bearing angles I r? 1 1 MW]

[0084] When the radar operates with a carrier frequency change from one recurrence to another over the observation time At, the observations are mutually independent and the sequence of estimated bearing angles JA 1 is random.

[0085] It can then be modeled by a normal N distribution with zero mean value and standard deviation for a target signal-to-noise ratio of zero (SNRC = 0) •

[0086] GN(0, O^(0))

[0087] In the particular case of high-resolution (metric or better) sea echoes, this modeling of the random sequence J p | can be refined taking into account the observed characteristics of the sea clutter. For example, a wave front may occasionally generate a sea clutter echo with a nearly fixed arrival angle over the observation period At. H may then become necessary to model these phenomena to design a more efficient detection process.

[0088] According to the general statistical theory of radars, the designer of a radar seeks to construct the detector maximizing the probability of detection of the target PD, for a constrained probability of false alarm Pf.

[0089] The Neyman-Pearson lemma, known to those skilled in the art, allows a detection test T to be constructed from a likelihood ratio.

[0090] The optimal detection test T is thus constructed:

[0091] > C

[0092] With S 'a Likelihood anointing of the observation sequence S under the hypothesis H.

[0093] The test is optimal in the sense that it maximizes power, in the statistical sense of this term, i.e. maximizing the probability PD subject to the constraint of the probability Pf.

[0094] A plot will be generated from the observations as soon as the test is validated, i.e. the ratio of the two likelihood functions greater than the suitably chosen threshold.

[0095] This optimal detection test is then worked on, in particular by expressing the likelihood functions as a function of the models made, until a sufficiently simple expression appears corresponding to a detection treatment which remains optimal, but which is simplified.

[0096] Approximations are then introduced on this simplified optimal test in order to implement it. The test actually applied then becomes suboptimal, but it corresponds to both the optimal test and the simplified optimal test.

[0097] This leads to the highlighting of new hybrid TFAC detection treatments in the sense that they rely on tests combining the elementary signals Sjy} from the sum channel and the multi-channel gap measurement signals QJ.

[0098] In the general case of mutually independent observations, the likelihood function can, for example, be rewritten as follows: [°°"1 A(SjH)=n^P(SiGj|H)

[0100] With SvgJW) 'a probability density of the random variable under the hypothesis H. The expression of this probability density depends on the modeling choices made for echoes from a target and those from noise.

[0101] With reference to [Fig.2], a particular example of a possible hybrid TFAC detection treatment is given.

[0102] We are considering the case of a progressive scanning radar.

[0103] The radar system antenna is subdivided into M+l sub-antennas.

[0104] In a step 110 of the process 100, the signal on the sum channel Svj is obtained at time i by summing the signals received by each of the sub-antennas.

[0105] In parallel, one or more signals on the difference channel are obtained at time i. They are denoted s^, with k an integer between 1 and K.

[0106] In the case where M is equal to 1, the signal on the difference channel is obtained by taking the difference between the signals received by the first sub-antenna and the second sub-antenna.

[0107] In the case where M is greater than 2, we no longer speak of a difference channel. We have M+l sub-antennas providing M+l signals at time i, which must be combined in a suitable manner (functions F & on [Fig. 2]) to obtain K signals "on the difference channel" in order to estimate, in combination with the signal on the sum channel, the angle of arrival. Several techniques, known to those skilled in the art, can be used to combine the signals provided by the sub-antennas.

[0108]

[0109]

[0110] [YES]

[0112]

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121]

[0122]

[0123] It should be noted that, in the case of an antenna with more than two channels, we simply have more information to estimate the angle of arrival, which remains a scalar quantity. In step 120, a deviation calculation is performed. From the signal on the sum channel and the signal(s) on the difference channel, an arrival angle Q is estimated. The next step, 130, involves the implementation of the hybrid detection processing. One possible treatment is as follows. Assuming the presence of a target C visible over the observation period At, we obtain a time sequence S^G of hybrid instantaneous signals S^G1, each hybrid instantaneous signal associating an instantaneous signal of sum channel s^j)ct an instantaneous signal of deviation Q j'j, where i indexes a box according to the direction in recurrence (or azimuth or time) and j indexes a box according to the direction in distance. In Figure 3, a target is present at the center of the observation domain. The top image corresponds to the input and the middle image to the input Q.Çjy. We observe in particular that the estimated angle of arrival evolves progressively for recurrences in the vicinity of the center of the image. In a first post-integration substep - PI of the signal on the sum channel, 132, the instantaneous signals Sjj on the sum channel are summed quadratically along the time axis (i.e. the index i): A'-l y— 2 With j being the cell distance index and q a summation integer. Process 100 includes a second sub-step of filtering the deviation signal 134. Given the equation modeling the arrival angles Q in the presence of a target: Gi = Azc-Az0-V^^At+Ui It is shown that the expression for the time filter h adapted for searching for a target with a fixed azimuth over the duration Ai is as follows: h(i) = V Bal.-^At Assuming N is odd for simplicity, the so-called compressed arrival angle Gconîp^j) (for the distance cell j and the recurrence i) at the end of the filtering is then obtained by convolution on the adjacent recurrences: comp,i(j) ai &i+q (j)'h(N-(i+C[) + 1) y 2

[0124] The image at the bottom of Figure 3 represents the compressed arrival angle. comp'

[0125] It is observed that the echoes at the center of this image appear clearly above the ambient noise, illustrating the importance of taking into account the deviation from the target detection processing.

[0126] In a third dimension reduction substep, 136, the two signals are combined into a single scalar metric: SZPI and Qcomp compressed

[0127]

[0128] With a covariance matrix whose coefficients depend on the modeling choices made. This metric achieves dimensionality reduction.

[0129] In a fourth comparison substep 138, the value of obtained at the output of the substep is compared to a properly calculated hybrid detection threshold to respect the P fa of stress.

[0130] When this threshold is exceeded, a plot P is generated at the output of step 132 of hybrid TFAC processing.

[0131] Plot P is characterized by three-dimensional position information, with an angular accuracy that is that of the deviation and no longer that of the opening of the main lobe of the radar beam.

[0132] The process 100 is iterated. It can be iterated for each recurrence i (the "sliding" mode, preferred embodiment), or by batch of recurrences (the "non-sliding" mode).

[0133] Downstream of step 130 of hybrid TFAC processing, the process 100 may include a step 140 of plot dressing, allowing the use of information obtained from other treatments to complete the information of plot P. However, according to the invention, the outputs of the deviation measurement step are no longer used as input to the dressing step.

[0134] The invention thus proposes to go beyond conventional TFAC detectors, which rely on the search for abnormally powerful echoes, with a hybrid TFAC detector, which relies on the search for abnormally powerful and abnormally stable echoes in angle of arrival over a short time horizon much less than one second (At < ls).

[0135] If the preferred embodiment has been presented above in the case of processing a radar signal, the detection method according to the invention applies to the processing of a sonar signal.

[0136] It should be noted that, in the prior art, the so-called "track before detection" (TBD) methods are known, which start from the raw radar signal. to detect a moving target. Such a method takes the power of the received signal as input. In some embodiments, it can also take the deviation signal as input. However, to detect a target, this method samples the signal at a long sampling rate, on the order of a second, and observes for a long time, for several tens of seconds, precisely to be able to detect the target's movement. In contrast, the method according to the invention, which also takes the power of the received signal and the deviation signal as input, operates at a short sampling rate, on the order of a millisecond, and observes for a short time, a few tens of milliseconds and at most a hundred milliseconds. This short observation time allows the target to be considered stationary.

[0137] Moreover, as a detection method, the hybrid AC TF method according to the invention delivers detections or plots P.

[0138] On the contrary, a TBD process delivers leads.

Claims

Demands

1. A computer-implemented method (100) for detecting a target (C) of the constant false alarm rate (TFAC) type for a multi-channel detection system, characterized in that the method, considering an observation period short enough for any target to be considered stationary, implements, for each sampling step (j) along a distance direction (D) of the multi-channel detection system, a detection test on a sequence of hybrid signals along a recursive direction of the multi-channel detection system, the hybrid signal sequence covering the observation period and each hybrid signal associating a sum channel signal (s23) and a deviation signal (£.) for each sampling step (i) along the recursive direction, the method generating a target detection when the detection test is verified.

2. A method according to claim 1, wherein the detection test corresponds to an optimal test T^G, the optimal test being written: c with A a likelihood function of the time sequence S of hybrid instantaneous signals either under the assumption Ho of the presence of a target C or under the assumption of the absence of a target, and Sjj a hybrid detection threshold, the method leading to the generation of a detection, or plot, when the detection test is verified, the hybrid detection threshold being calculated to respect a constraint false alarm probability.

3. A method according to claim 2, comprising, for the case of a multi-channel scanning detection system, a filtering step (134) for calculating a compressed arrival angle GCOIup from the deviation signal according to the relation: Gcompjtj) ~ i+q( .7 ) ~ Ù + Q) + 1) with h a time-domain filter adapted for searching for a fixed target over the transit time of a main lobe of an antenna of the multi-channel detection system over the target, N the number of signals hybrids in the sequence, i a sampling step index along the direction in recurrence, j a sampling step index along the direction in distance, and q an integer summation index.

4. Method according to claim 3, comprising a post-integration step (132) of the sum channel signal to obtain a compressed sum channel signal S^pj : Szpij U) ^q=.^ 1 szi+q ( J ) 12 With N the number of hybrid signals in the sequence, i a sampling step index along the recurrence direction, j a sampling step index along the distance direction, and q an integer summation index.

5. A method according to claim 4, comprising a dimension reduction step (136) for combining the compressed sum track signal S-^pj and the compressed deviation signal Gconip according to a predefined metric to obtain a scalar value, and a step for comparing the scalar value to the hybrid detection threshold

6. Method according to claim 5, wherein the metric D^^a is defined by: 2 H ( 4 N 1 S'ZPlJJ) ^Maha,i J ~ 1 ^TRT / a with ^ppp a covariance matrix,

7. A method according to any one of the preceding claims, wherein the method enables the search for abnormally strong and abnormally stable echoes in angle of arrival over a short observation period, less than one second, preferably equal to 10ms, and a short sampling step, less than ten milliseconds, preferably equal to 1ms.

8. Multi-channel detection system incorporating a computer programmed to perform the steps of the process according to any one of claims 1 to 7.

9. System according to claim 8, the system being of the radar type or of the sonar type.

10. Computer program comprising program code instructions for executing the steps of the process according to any one of claims 1 to 7 when said program is executed on a computer of a multi-channel detection system according to claim 8 or claim 9.

Images