JPH0843520A - Pulse-signal sorting apparatus - Google Patents

Abstract

PURPOSE:To obtain a pulse-signal sorting apparatus by which a sorting operation can be performed automatically and at a high speed by increasing or decreasing the number of sorting operations and sorting pulse signals into the proper number of sorting operations. CONSTITUTION:The feature amount of respective radar pulses contained in a mixed-radar-pulse file 1 is found by a feature-amount extraction part 2. Then, a cluster-number selection part 38 sets the number of clusters as N, e.g. 1, and the radar pulses are sorted into the number of clusters by a sorting part 90. When the number of feature amounts is set as P, the feature amount of the respective radar pulses is plotted in one point in a P-dimensional feature space. The feature amount of all radar pulses is plotted in the feature space, plots at close distances are collected, and the set of the collected radar pulses is regarded as a cluster. Then, an irregularity in respective clusters is found by a cluster-irregularity computation part 20. When the irregularity is large, the cluster is divided by a cluster division part 35. When the irregularity is small, it is regarded as a sorted result. Thereby, one number or divided patterns is reduced, and an optimum sorting operation can be reached at a high speed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pulse signal classifying device for receiving a pulse train in which pulse signals repeatedly transmitted in time from a plurality of signal sources are mixed and classifying the pulse train into pulse trains for each signal source. .

[0002]

2. Description of the Related Art When receiving a pulse train in which pulse signals repeatedly transmitted in time from a plurality of signal sources are mixed and classified into pulse trains for each signal source, conventionally, as shown in FIG. The pulse feature amounts of are extracted and classified using the histogram of these feature amounts. Figure 21 shows D.C.
urtis Schleher ■ Introduction to Electronic Warfa
re ■ pp.47-57 It is an analogy from Artech House 1986.

The operation of the conventional example shown in FIG. 21 will be described. A figure in which the feature quantity of each radar pulse included in the mixed radar pulse file 1 is extracted by the feature quantity extraction unit 2 and the feature quantity extracted by the histogram creation unit 10 is taken on each axis.
Create a histogram like 22.

Next, an area surrounded by a portion whose histogram value is 0 is extracted, and the pulse signals included in this area are set as one set. In this example, when the values of the feature amounts of the two pulse signal sources are close to each other and are adjacent to each other in the histogram, it cannot be distinguished as two sets, and the number of sets to be classified is wrong. There was a problem.

If it is not known in advance how many sets will be classified, it has conventionally been necessary to input a threshold value or the like in advance in order to determine the number of classifications, as shown in the overall configuration diagram of FIG. This FIG. 23 is shown in JP-A-63-257.
It is inferred from No. 798, "Standard Pattern Creation Method".

The operation of the conventional example shown in FIG. 23 will be described. The feature amount of each radar pulse included in the mixed radar pulse file 1 is extracted by the feature amount extraction unit 2, and the threshold value τ input unit
Enter the value of τ at 80. In the cluster number selection unit 38, the number of clusters is selected to be 1, all data are considered as one cluster, and in the cluster data distance and the threshold τ comparison unit 82, if there is a cluster having a data distance equal to or greater than the threshold τ. , The cluster is divided by the cluster dividing unit 35, and the data distance in each cluster and the threshold τ comparison unit are returned to. On the other hand, if the distances of the data of all combinations in each cluster are equal to or smaller than the threshold value τ, this is set as the classification result 4. Thus, it was necessary for a human to determine the threshold τ.

Further, when the radar pulse is classified, FIG.
As shown in the overall configuration diagram of No. 3, although the classification is performed only by the reception time, the correction is not performed by using the reception time after classifying by the other feature amount. FIG. 24 is inferred from "Pulse Congestion Separation Circuit" of Japanese Utility Model Laid-Open No. 61-113428.

Next, the operation will be described. In FIG. 24, the radar pulse included in the mixed radar pulse file 1 is determined by the period extraction unit 91 to obtain the radar pulse transmission period T of one radar, and the radar pulse received at this period T is extracted by the pulse extraction unit 97. To do. By repeating this, the radar pulse train for each radar is classified, and the classification result 4 is obtained.

[0009]

Since the conventional pulse signal classifying apparatus is constructed as described above, it is necessary to give an index for carrying out some kind of classification, such as a threshold value, in order to carry out automatic classification. However, it was difficult to set this as a fixed value in advance. Also,
When the pulse reception time has regularity, the regularity of pulse reception time was not used to perform classification correction.

The present invention has been made in order to solve the above-mentioned problems and is to automatically determine the number of classifications and to perform classification correction using pulse reception time.
The purpose is to perform these classifications at high speed.

[0011]

According to a first aspect of the present invention, there is provided a pulse signal classification device which receives a pulse train of pulse signals repeatedly transmitted from a plurality of signal sources, and which receives the pulse train from each of the signal sources. The pulse signal classifying device for recognizing and classifying the corresponding pulse train is provided with a unit for classifying the pulse signal into an appropriate class number by increasing or decreasing the class number, which is the number of sets to be classified.

A pulse signal classifying apparatus according to a second aspect of the present invention is a means for obtaining a feature quantity estimation error from a receiver measurement accuracy file, and increasing / decreasing the number of classifications based on the obtained feature quantity estimation error to make an appropriate classification. It is provided with a means for classifying pulse signals into numbers.

A pulse signal classification device according to a third aspect of the present invention includes a means for obtaining a feature quantity estimation error from a receiver measurement accuracy file, and a certain classification based on the obtained feature quantity estimation error and the variation of the feature quantity of the classification result. It is provided with a means for calculating an evaluation of the classification result when the pulse signal is classified into numbers, and a means for classifying the pulse signal into a classification number that makes the evaluation appropriate by increasing or decreasing the classification number.

According to a fourth aspect of the pulse signal classifying device, a means for obtaining a feature quantity estimation error from a receiver measurement accuracy file, and a feature quantity of each pulse signal is normalized using the obtained feature quantity estimation error. And a means for calculating the evaluation of the classification result when the pulse signal is classified into a certain number of classifications from the variation of the feature amount of the classification result, and a classification for increasing or decreasing the number of classifications to make the evaluation appropriate. It is provided with a means for classifying pulse signals into numbers.

According to the fifth aspect of the present invention, the pulse signal classifying apparatus classifies the pulse signals into a certain number of classifications on the basis that the variation of the feature amount of the classification result indicates that the variation of the pulse signal is the same for each signal source. And means for calculating the evaluation result of the classification result, and means for classifying the pulse signal into the number of classifications for increasing or decreasing the number of classifications to make the evaluation appropriate.

The pulse signal classification device according to the invention of claim 6 calculates the evaluation of the classification result when the pulse signal is classified into a certain number of classifications from the assumed feature quantity estimation error and the variation of the feature quantity of the classification result. Means, a means for classifying the pulse signal into the number of classifications that makes the above evaluation appropriate by increasing or decreasing the number of classifications, and a means for correcting the feature quantity estimation error assumed from the variation of the feature quantity of the classification result. It is a thing.

The pulse signal classification device according to the invention of claim 7 comprises means for obtaining a pulse interval, and means for correcting the classification result based on the obtained pulse interval.

The pulse signal classifying apparatus according to the invention of claim 8 comprises means for correcting the classification of the duplicated pulses in which the pulses from a plurality of signal sources are simultaneously received.

A pulse signal classification device according to a ninth aspect of the present invention comprises means for estimating a pulse repetition interval from a set of classified pulses, and means for correcting the classification result using the estimated pulse repetition interval. Is.

According to a tenth aspect of the present invention, in a pulse signal classification device, when classifying a pulse signal into several sets, a means for classifying the degree of belonging to each set by an ambiguous value and classifying the pulse intervals. A means for determining and a means for correcting the classification result based on the obtained pulse interval by using the degree of belonging as a weighting value for classification correction are provided.

The pulse signal classifying apparatus according to the invention of claim 11 comprises means for integrating the classification result.

A pulse signal classification device according to a twelfth aspect of the present invention comprises means for extracting discrete values of the pulse waveform, and performs classification using the discrete values of the pulse waveform as the feature quantity for classification. Is.

The pulse signal classification device according to the thirteenth aspect of the present invention comprises means for giving an initial value of the classification when starting the classification.

According to a fourteenth aspect of the present invention, there is provided a pulse signal classification device for storing the characteristic amount of a pulse signal of a known signal source, and the representative value of the pulse characteristic amount of each pulse train is the value of the stored characteristic amount. It is equipped with a means for classifying that it is taken only.

[0025]

According to the invention of claim 1, the pulse signal is classified into an appropriate number of clusters by means of increasing or decreasing the number of classifications to classify the pulse signal into an appropriate classification number.

According to the second aspect of the present invention, the cluster size is adjusted based on the feature value estimation error.
Classify the pulse signal into an appropriate number of sets.

According to the third aspect of the present invention, whether or not the classification is appropriate is evaluated from the estimation error of the characteristic amount and the variation of the characteristic amount of the classification result.

According to the invention of claim 4, the feature quantity estimation error is obtained from the receiver measurement accuracy file, and the feature quantity of each pulse signal is normalized using the obtained feature quantity estimation error.
From the variation of the feature amount of the classification result, the evaluation of the classification result when the pulse signal is classified into a certain classification number is calculated, and the pulse signal is classified into the classification number that makes the above evaluation appropriate by increasing or decreasing the above classification number. To do.

According to the fifth aspect of the present invention, it is assumed that the pulse signals have the same variation for each signal source based on the variation of the feature amount of the classification result, and the classification result is evaluated when the pulse signals are classified into a certain number of classifications. Is calculated and the number of classifications is increased or decreased to classify the pulse signal into the number of classifications that makes the above evaluation appropriate.

According to the sixth aspect of the present invention, the evaluation of the classification result when the pulse signal is classified into a certain classification number is calculated from the assumed feature amount estimation error and the variation in the characteristic amount of the classification result, and the above classification number is calculated. The pulse signals are classified into the number of classifications that increases and decreases to make the above evaluation appropriate, and the feature quantity estimation error assumed from the variation of the feature quantity of the classification result is corrected.

According to the invention of claim 7, the pulse interval is obtained from the pulse arrival time of the pulses classified into the same cluster by the pulse interval obtaining means.
The pulse which is erroneously classified is recognized, and the classification result is corrected by the means for correcting the classification result based on the obtained pulse interval.

According to the eighth aspect of the present invention, in the means for correcting the classification of the duplicated pulses which have received the pulses from the plurality of signal sources at the same time, when the plurality of pulses are received as one pulse in duplicate, the pulse is received. Is a pulse whose characteristic amount is significantly different from other pulses, and is thus recognized as a duplicate pulse, and the classification result is corrected by recognizing that two or more pulses have been received at this duplicate pulse reception time.

According to the ninth aspect of the present invention, the pulse repetition interval is estimated from the pulse having the characteristic amount close to the average value of the pulse characteristic amount of each pulse train by the means for estimating the pulse repetition interval from the set of classified pulses, Correct the classification result from the pulse repetition interval.

According to the tenth aspect of the present invention, when the pulse signals are classified into several sets, the degree of belonging to each cluster is represented by an ambiguous value indicating the degree of belonging to each set. Then, the above-mentioned degree of belonging is used as a weighting value for classification correction, and the classification result is corrected based on the obtained pulse interval, whereby the classification result is corrected more accurately.

In the eleventh aspect of the present invention, the accuracy of the feature amount obtained from the pulse train of the classification result is increased by integrating the classification results of several times.

According to the twelfth aspect of the present invention, means for extracting the discrete values of the pulse waveform is provided, and more accurate classification is performed by using the discrete values of the pulse waveform as the classification feature amount.

According to the thirteenth aspect of the present invention, there is provided means for giving an initial value of the classification when starting the classification, and by giving an initial value close to the finally obtained answer of the classification,
Fast classification.

According to the fourteenth aspect of the present invention, there is provided means for accumulating the characteristic amount of the pulse signal of the known signal source, and the representative value of the pulse characteristic amount of each pulse train is only the value of the accumulated characteristic amount. Since classification is performed, classification is performed at high speed.

[0039]

【Example】

Example 1. The present invention will be described with reference to the drawings. 1 is an overall configuration diagram showing an embodiment of a pulse signal classification device according to the invention of claim 1, and FIG. 2 is an example of a simultaneous distribution of a plurality of feature quantities of a radar pulse signal.

Next, the operation will be described. In this example, when radar pulses transmitted from a plurality of radars are received, these radar pulses are classified for each radar.
First, the feature amount extraction unit 2 obtains the feature amount of each radar pulse included in the mixed radar pulse file 1.

Next, the cluster number selecting unit 38 sets the number of clusters to N, for example, 1 and classifies the radar pulse into this cluster number by the classifying unit 90 which classifies the radar pulse into the given number of clusters. When the number of feature quantities is P, the feature quantity Xn of each radar pulse is plotted at one point in the P-dimensional feature space. The feature quantities of all radar pulses are plotted in this feature space, plots with close distances are put together, and the set of the put together radar pulses is taken as a cluster.

As a method of collecting the plots, there is, for example, the K-means method. Clusters are K1 and K2 respectively
..., KL, its center is Mi, and the distance between Xn and Mi is ||
If Xn-Mi ||, the K-means method is

[0043]

[Equation 1]

The classification is performed so as to minimize.

Next, the cluster variation calculation unit 20 obtains the variation of each cluster, and if this variation is large,
The cluster dividing unit 35 divides the cluster, and if it is smaller, this is set as the classification result 4.

Example 2. FIG. 3 is an overall configuration diagram showing an embodiment of the pulse signal classification device according to the invention of claim 2.

Next, the operation will be described. In this example, pulse signals transmitted from a plurality of radars are received, and when the number of radars is unknown, these radar pulses are classified for each radar.

First, the characteristic amount of each radar pulse included in the mixed radar pulse file 1 is obtained by the characteristic amount extraction unit 2. The cluster number selection unit 38 sets the number of clusters to 1.

From the receiver measurement accuracy file 31, the feature quantity estimation error calculation unit 32 calculates the estimation error of each feature quantity. This receiver measurement accuracy file records the measurement accuracy of the receiver.If you also record temperature characteristics and frequency characteristics in addition to this, measurement error will occur for each temperature and frequency of the target signal. Can be obtained, and a more accurate measurement error can be estimated.

Intra-cluster data distance and estimation error comparison unit
In 37, if the difference in the value of the feature amount of each pulse data in each cluster is larger than the feature amount estimation error, this cluster is divided by the cluster division unit 35, and the data distance in each cluster and the estimation error comparison unit are divided. Return to 37. If it is smaller, classification is completed and classification result 4 is obtained.

As a result, the classification can be automatically performed even if the number of radars transmitting radar pulses is unknown.

Example 3. FIG. 4 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 3.

Next, the operation will be described. In this example, radar pulses transmitted from a plurality of radars are received, and when the number of radars is unknown, these radar pulses are classified for each radar.

Similar to the second embodiment, the feature amount of each radar pulse is obtained, and the cluster number selection unit 38 sets the number of clusters to 1
And select. Classifier 90 for classifying given number of clusters
Then, sort into the selected number of clusters. For example, the classification can be performed by the K-means method described in the first embodiment.

This classification result is a temporary classification result 44.
An evaluation value is calculated by the evaluation value calculation unit 42 from the squared error ε of the temporary classification result and the estimation error calculated in the same manner as in the second embodiment. The evaluation value J is, for example, the estimation error of each feature amount that is σ _P Pp
(1 ≦ p ≦ P), | Ki | is the number of pulse signals included in the cluster Ki, Xnp is the p-th feature quantity of the pulse signal Xn, Mi
Let p be the p-th feature of cluster Ki, and use AIC (Akaike
Information Criterion)

[0056]

[Equation 2]

Is calculated as Next, the cluster number increasing unit 43
Increase the number of clusters by. And classify in the same way,
Obtain the evaluation value of the classification. The cluster number determination unit 45 compares the evaluation value of the previous cluster number, and if the evaluation value is smaller, the cluster number is further increased by 1 and the process is continued. On the contrary, if the evaluation value seems to be large,
The classification result by the number of clusters for which the minimum evaluation value was obtained,
The final classification result is 4. In practice, it is conceivable that the evaluation value is calculated with all the cluster numbers L (L is the number of data or less) and the number of clusters that gives the minimum evaluation value is searched.

An example in which pulse signals are classified by this method will be shown.

It is assumed that a radar pulse feature amount as shown in FIG. 5 is obtained from a radar pulse train in which radar pulses from two radars are mixed. (The radar number is shown for reference only and is not known in advance.)

FIG. 6 shows evaluation values obtained as a result of classification by setting the radar pulse having the characteristic amount of FIG. 5 to each cluster number. From this result, it can be seen that the correct number of clusters is obtained. The result of classifying with this correct number of clusters is
As shown in FIG. 7, it can be seen that they are correctly classified.

FIG. 8 shows the result of obtaining the evaluation value by adding the feature amount of the direction to the feature amount of FIG. 2 and classifying the number of clusters into three dimensions, and it can be seen that the correct number of clusters is obtained. . The result of classification with the correct number of clusters is shown in FIG. 9. Since the distributions of the feature quantities are mixed, the two pulses are erroneously classified, but the other pulses are correctly classified.

Example 4. FIG. 10 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 4.

Next, the operation will be described. In this example, radar pulses transmitted from a plurality of radars are received, and when the number of radars is unknown, these radar pulses are classified for each radar.

Similar to the second embodiment, the feature quantity of each radar pulse is obtained, and the feature quantity is characterized by the estimation error of each feature quantity obtained by the feature quantity estimation error calculation section 32 as in the second embodiment. The amount normalization unit 40 normalizes. Using the normalized feature quantity, the classification unit 90 that classifies the cluster into a given number of clusters performs classification with each number of clusters, as in the third embodiment.
The evaluation value calculation unit 42 obtains the evaluation value of the classification result for each number of clusters, and the cluster number determination unit 45 obtains the optimum number of clusters.

However, unlike the third embodiment, the function of the estimation error (σp in the third embodiment) is not necessary when obtaining the evaluation value.

[0066]

(Equation 3)

It becomes

Example 5. FIG. 11 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 5.

Next, the operation will be described. In this example, when radar pulses transmitted from a plurality of radars are received and these radar pulses are classified into a radar pulse train for each radar, if there is a possibility that the radar changes the feature amount, The variation in the feature amount in 1 is not due only to the estimation error of the feature amount, and the classification result cannot be evaluated in the third and fourth embodiments. Therefore, it is considered that the variation in the feature amount is the same in each cluster. Is an example of performing.

Similar to the third embodiment, the processing up to the classifying unit 90 for classifying into the given number of clusters is performed, and the evaluation value calculating unit 42 obtains the evaluation value of the classification for each number of clusters. This evaluation value J differs from the third embodiment in that, from the squared error of this classification result and the number of pulses | Ki | included in each cluster, for example,

[0071]

[Equation 4]

Is calculated as Then, similarly to the third embodiment, the classification result with the number of clusters for which the minimum evaluation value is obtained is set as the final classification result 4.

Example 6. FIG. 12 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 6.

Next, the operation will be described. This example is an example of receiving radar pulses transmitted from a plurality of radars and classifying these radar pulses for each radar when the receiver measurement accuracy is unknown.

Similar to the third embodiment, the processes up to the classifying unit 90 for classifying into the given number of clusters are performed. However,
Since the receiver measurement accuracy is unknown, the feature estimation error of the radar pulse cannot be obtained. Therefore, the feature quantity estimation error setting unit 92 assumes the feature quantity estimation error. Thereafter, the classification result with the optimum number of clusters is obtained in the same manner as in the third embodiment.

From this classification result, the variation in the feature amount of each cluster is obtained. Here, the variation is represented by the standard deviation. If the assumed feature amount estimation error (this error is also represented by standard deviation) is compared, and the assumed feature amount estimation error is larger or smaller than the variation of each cluster (for example, more than twice If it is less than half, the feature quantity estimation error correction unit 93 replaces the feature quantity estimation error with the variation of the feature quantity of each cluster, and the cluster number correction unit 95 returns the number of clusters to 1.

After this processing, the above-described processing for obtaining the optimum number of clusters is repeated, and if the assumed feature quantity estimation error and the variation of each cluster have the same magnitude, this is regarded as the classification result. By doing so, classification can be performed even if the feature quantity estimation error is not known.

Example 7. FIG. 13 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 7.

Next, the operation will be described. In this example, when radar pulses transmitted from a plurality of radars are received and these radar pulses are classified into radar pulse trains for each radar, the regularity of pulse reception time is used to correct the classification result. Is.

The result of classification by some sort means is defined as a temporary classification result 44. In this embodiment, the classification of the first embodiment is performed.

When the distributions of the feature quantities overlap as shown in FIG. 2, some pulses may be misclassified. However, if it is a radar pulse, frequency jitter
Even if staggered, it is received with a certain degree of temporal periodicity or regularity. Therefore,
The pulse interval estimation unit 50 obtains the radar pulse reception interval of each cluster, and the greatly different pulse interval extraction unit 51 extracts a pulse interval extremely different from other pulse intervals.

It is considered highly probable that the two pulses before and after the extracted pulse interval constituting the pulse interval are erroneously classified, and the classification and correction unit 52 using the pulse interval divides these pulses into other clusters. Move, find the pulse interval, and if this is similar to other pulse intervals,
The classification is corrected so that this pulse is moved. All the pulses forming the greatly different pulse intervals are moved and examined, and the result of correcting the classification is set as the classification result 4. As a result, it is possible to obtain a more accurate classification result than the temporary classification result.

Example 8. FIG. 14 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 8.

Next, the operation will be described. In this example, when radar pulses transmitted from a plurality of radars are received and these radar pulses are classified into a radar pulse train for each radar, the characteristic amount of the overlappingly received pulses is the characteristic of other pulses. This is an example in which the classification result is corrected by using the fact that it is located far from the feature space as compared with the quantity.

The result of classification by some sort of classification means is defined as the temporary classification result 44. When radar pulses from a plurality of radars are partially overlapped and received, a plurality of pulses should be received, but only one radar pulse is received. The feature amount of this pulse has a feature amount significantly different from the feature amounts of other pulses.

If the temporary classification result 44 has a cluster composed of one pulse, the pulses included in this cluster are highly likely to be the pulses received in duplicate. Therefore, in the classification correction unit 55 using the overlapping pulse, if this cluster is erased and the pulse interval becomes more regular when considering that there is a pulse at the time when this pulse is received in another cluster, this time Then, assuming that a pulse having the characteristic value of the representative value of the cluster is received, the classification result is corrected. As a result, accurate classification can be performed even if radar pulses are received in duplicate.

Example 9. FIG. 15 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 9.

Next, the operation will be described. In this example, when radar pulses transmitted from a plurality of radars are received and these radar pulses are classified into a radar pulse train for each radar, the pulse repetition interval of each cluster is obtained and this pulse repetition interval is used. This is an example of correcting the classification result by

Even if the distributions of the feature quantities are mixed as shown in FIG. 2, there is a high possibility that the clusters near the center of each cluster are correctly classified, and that the clusters are misclassified toward the periphery. Is high. Therefore, the pulse repetition interval estimation unit 56 obtains the pulse repetition interval of each radar as the greatest common divisor of the reception intervals of pulses near the center of each cluster. The pulse arrival time estimation unit 57 estimates the pulse reception time of each radar from this pulse repetition interval.

The comparison unit 59 between the estimated pulse arrival time and the actual pulse arrival time makes a comparison, and if the pulse arriving at a time greatly apart from the estimated pulse arrival time is a pulse of another cluster, the pulse Move with the moving unit 54,
to correct. This correction is repeated, and the result of correcting all the pulses arriving at a time greatly apart from the estimated pulse arrival time is set as a classification result 4.

Example 10. FIG. 16 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 10.

Next, the operation will be described. In this example, when receiving radar pulses transmitted from a plurality of radars and classifying these radar pulses into a radar pulse train for each radar, the regularity of the reception time of the radar pulse and the classification result of fuzzy clustering This is an example of correcting a classification result by using an ambiguous expression, and is an example in which the classification method of the seventh embodiment is changed to fuzzy clustering.

When the distributions of the feature quantities are mixed as shown in FIG. 2, there is a high possibility that the clusters near the center of each cluster will be correctly classified, and there will be a possibility that the clusters will be misclassified toward the periphery. high.

When the classification is performed by fuzzy clustering, the degree of belonging of each pulse to each cluster is obtained as a real value between 0 and 1, and this is used as a degree of belonging file 61 of each pulse to each cluster. When the classification result is corrected by the pulse interval as in the seventh embodiment, since the pulse interval is obtained from two pulses, it is necessary to check which pulse makes the wrong classification. Therefore, it is considered that the pulse having a lower degree of belonging in the degree of belonging file 61 of each pulse to each cluster is wrongly classified, and the classification correcting unit 52 using the pulse interval assigns each pulse to each cluster. Correct the classification based on the degree file 61.

Example 11. FIG. 17 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the invention of claim 11.

Next, the operation will be described. In this example, when radar pulses transmitted from multiple radars are received and these radar pulses are classified into radar pulse trains for each radar, the classification results are added together to generate the center of each cluster of the classification results. The value is obtained accurately.

When the pulse signal source outputs a signal for a relatively long time, the time required to accumulate data for classification by comparison with that is short, and when the classification is repeated many times, the same pulse as the previous classification result is obtained. Considering that there is a high possibility that the signal source is outputting a signal, the classification result integrating unit 63 adds up the classification results several times.

If the pulse signals from the same signal source are included in the classification results of several times, the number of pulses included in each cluster is increased by adding the classification results together, and The center is closer to the true value.

Example 12. FIG. 18 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the twelfth aspect of the invention.

Next, the operation will be described. In this example, when receiving radar pulses transmitted from a plurality of radars and classifying these radar pulses into a radar pulse train for each radar, by using discrete values of pulse waveforms as a feature amount for classification, It is intended to classify using all the features included in the pulse waveform.

The radar pulse waveform included in the mixed radar pulse file 1 is converted into discrete values by the pulse waveform discrete value extraction unit 70. As it is, when extracting the part of the pulse signal from the observed signal, it is possible that the pulse signal pattern is misaligned by cutting out to the excess part or cutting off part of the pulse signal, The feature conversion unit 71 performs Fourier transform or wavelet transform to convert into a feature amount that is not affected by the positional deviation.
(However, if the signal has a good S / N, it is not necessary to perform feature conversion.)

If classification is performed using this feature, classification using all features included in the pulse signal can be performed. In this embodiment, a classification unit that classifies into a given number of clusters
Classification is performed at 90, and classification result 4 is obtained.

Example 13. FIG. 19 is an overall configuration diagram for explaining an embodiment of the pulse signal classification device according to the thirteenth invention.

Next, the operation of the thirteenth embodiment will be described. In this example, when radar pulses transmitted from a plurality of radars are received and these radar pulses are classified into a radar pulse train for each radar, initial cluster division is set in order to perform classification at high speed. In this embodiment, the initial cluster division is set in the classification of the third embodiment.

A classification algorithm such as the K-means method first divides the cluster into appropriate clusters, and changes the division of the cluster so that the evaluation value is the best, even if the evaluation value is good even if the division is changed. When it disappears, it is considered that the classification result has been obtained. Therefore, if the first cluster division is close to the classification result, the classification ends faster.

Therefore, the pulse signal source outputs a signal for a relatively long time, and the time for accumulating data for classification by comparison therewith is short, and when repeating classification many times, the same pulse as the previous classification result is obtained. Considering that the signal source is outputting a signal, the initial cluster division setting unit 74 sets and assigns the initial cluster division assuming that a cluster is formed at the feature of each pulse signal source in the previous classification result file 73. The classification of the classification unit 90 for classifying into the determined number of clusters is started. The subsequent processing is the same as in the third embodiment.

As a result, since many pulse signal sources are the same as the previous ones, the optimum classification result is obtained without changing the classification so much and the classification is completed at high speed.

Example 14. 20 is an overall configuration diagram for explaining an embodiment of a pulse signal classification device according to the fourteenth aspect of the present invention.

Next, the operation will be described. In this example, when radar pulses transmitted from a plurality of radars are received and these radar pulses are classified into a radar pulse train for each radar, the representative value of the feature amount of each cluster is This is an example in which classification is performed assuming that only known values are taken, and this embodiment is an example in which the representative value of the feature amount of each cluster takes only known values in the classification of the third embodiment.

It is assumed that the target radar characteristic amount file 75 includes all the characteristic amounts of the target radar. In this case, the characteristic amount of the representative value of the cluster obtained as a result of the classification is the value included in the target radar characteristic amount file 75. Therefore, when the cluster is divided, the representative value of each cluster can be considered to be the value included in the target radar feature amount file 75.

A classification algorithm such as the K-means method first divides into appropriate clusters, and changes the division of the clusters so that the evaluation value becomes the best even a little. Therefore, when changing this division, , It is considered that the representative value of the feature amount of the cluster takes only the known value recorded in the target radar feature amount file 75. In the normal classification, the center of the cluster can take an arbitrary value, but can take only a predetermined value, so that the pattern of division is reduced and the optimally considered division is reached at high speed.

[0112]

As described above, according to the invention of claim 1, the pulse signals can be classified more accurately.

According to the second aspect of the present invention, it is not necessary to determine the number of clusters, so that it can be completely automated.

According to the invention of claim 3, the optimum number of clusters can be automatically determined by a statistical method.

According to the invention of claim 4, the optimum number of clusters can be determined by a statistical method, and the feature quantity is normalized, so that the values at the time of classification are of the same order. Easy to calculate.

According to the invention of claim 5, the optimum number of clusters can be determined.

According to the invention of claim 6, the optimum number of clusters can be determined even if the feature quantity estimation error is not known.

According to the seventh aspect of the invention, the classification can be performed in consideration of the pulse arrival time intervals as well, and the classification can be performed more accurately.

According to the invention of claim 8, it is possible to prevent pulse omission due to simultaneous reception of pulses.

According to the ninth aspect of the invention, since the pulse arrival time can be estimated, it becomes possible to classify more accurately.

According to the tenth aspect of the present invention, when the classification correction is performed, the correction can be performed more accurately.

According to the eleventh aspect of the present invention, the characteristic amount of each radio wave source obtained from the classification result can be obtained more accurately.

According to the twelfth aspect of the invention, classification can be performed more accurately.

According to the thirteenth aspect of the invention, classification can be performed at high speed.

According to the fourteenth aspect of the invention, the classification can be performed at a higher speed.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram according to a first embodiment of the present invention.

FIG. 2 is an example of a simultaneous distribution of feature quantities of pulse signals.

FIG. 3 is an overall configuration diagram according to a second embodiment of the present invention.

FIG. 4 is an overall configuration diagram according to a third embodiment of the present invention.

FIG. 5 is an example of a characteristic amount of a radar pulse.

6 is an evaluation value of a classification result for each number of clusters of the radar pulse of FIG.

FIG. 7 is a result of the radar pulse classification shown in FIG.

8 is an evaluation value of a classification result for each number of clusters of the radar pulse of FIG.

9 is a classification result of the radar pulse of FIG.

FIG. 10 is an overall configuration diagram according to a fourth embodiment of the present invention.

FIG. 11 is an overall configuration diagram according to a fifth embodiment of the present invention.

FIG. 12 is an overall configuration diagram according to a sixth embodiment of the present invention.

FIG. 13 is an overall configuration diagram according to a seventh embodiment of the present invention.

FIG. 14 is an overall configuration diagram according to an eighth embodiment of the present invention.

FIG. 15 is an overall configuration diagram according to a ninth embodiment of the present invention.

FIG. 16 is an overall configuration diagram according to Example 10 of the present invention.

FIG. 17 is an overall configuration diagram according to an eleventh embodiment of the present invention.

FIG. 18 is an overall configuration diagram according to a twelfth embodiment of the present invention.

FIG. 19 is an overall configuration diagram according to Embodiment 13 of the present invention.

FIG. 20 is an overall configuration diagram according to Embodiment 14 of the present invention.

FIG. 21 is an overall configuration diagram of a conventional example in which classification is performed using a peripheral distribution.

FIG. 22 is an example of a histogram in which two feature amounts are taken on each axis.

FIG. 23 is an overall configuration diagram of a conventional example in which a threshold is given to determine the number of classifications.

FIG. 24 is an overall configuration diagram of a conventional example of classification using the reception time of a radar pulse.

[Explanation of symbols]

1 mixed radar pulse file 2 feature amount extraction unit 4 classification result 10 histogram creation unit 11 region extraction unit surrounded by frequency 0 20 cluster variation calculation unit 31 receiver measurement accuracy file 32 feature amount estimation error calculation unit 35 cluster division Part 37 Data distance in each cluster and estimation error comparison part 38 Cluster number selection part 40 Feature amount normalization part 42 Evaluation value calculation part 43 Cluster number increase part 44 Temporary classification result 45 Cluster number determination part 50 Pulse interval estimation part 51 Very different Pulse interval extraction unit 52 Classification correction unit using pulse intervals 54 Pulse moving unit 55 Classification correction unit using overlapping pulses 56 Pulse repetition interval estimation unit 57 Pulse arrival time estimation unit 59 Estimated pulse arrival time and actual pulse arrival time Comparison unit 60 Fuzzy clustering unit 61 For each pulse Affiliation file to cluster 63 Classification integration part 70 Pulse waveform discrete value extraction part 71 Feature conversion part 73 Previous classification result file 74 Initial cluster division setting part 75 Target radar feature amount file 80 Threshold τ input part 82 Each cluster Inner data distance and threshold τ comparison unit 90 Classification unit for classifying into a given number of clusters 91 Period extraction unit 92 Feature amount estimation error setting unit 93 Feature amount estimation error correction unit 94 Feature amount estimation error and classification result comparison unit 95 Cluster Number Corrector 97 Pulse Extractor

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yoshio Kosuge 5-1-1 Ofuna, Kamakura-shi Electronic Systems Research Center, Mitsubishi Electric Corporation

Claims (14)

[Claims] 1. A pulse signal classification device for receiving a pulse train composed of pulse signals repeatedly transmitted in time from a plurality of signal sources, and recognizing and classifying the pulse train corresponding to each of the signal sources. A pulse signal classification device comprising means for classifying pulse signals into an appropriate classification number by increasing / decreasing the classification number which is the number of sets. 2. A means for obtaining a feature quantity estimation error from a receiver measurement accuracy file, and a means for increasing / decreasing the number of classifications based on the obtained feature quantity estimation error and classifying the pulse signal into an appropriate number of classifications. The pulse signal classification device according to claim 1. 3. A means for obtaining a feature quantity estimation error from a receiver measurement accuracy file, and a classification result when a pulse signal is classified into a certain classification number based on the obtained feature quantity estimation error and the variation of the feature quantity of the classification result. 2. The pulse signal classification device according to claim 1, further comprising means for calculating the evaluation of 1., and means for classifying the pulse signal into a classification number that makes the evaluation appropriate by increasing or decreasing the classification number. 4. A means for obtaining a feature quantity estimation error from a receiver measurement accuracy file, a means for normalizing the feature quantity of each pulse signal using the obtained feature quantity estimation error, and a variation in the feature quantity of the classification result. From the above, a means for calculating the evaluation of the classification result when the pulse signal is classified into a certain classification number, and a means for classifying the pulse signal into the classification number for increasing or decreasing the classification number to make the evaluation appropriate are provided. The pulse signal classification device according to claim 1. 5. A means for calculating an evaluation of the classification result when the pulse signals are classified into a certain number of classifications, on the basis that the variation of the pulse signal is the same for each signal source, based on the variation of the feature amount of the classification result. The pulse signal classification device according to claim 1, further comprising means for classifying the pulse signal into a classification number that makes the evaluation appropriate by increasing or decreasing the classification number. 6. A means for calculating an evaluation of a classification result when a pulse signal is classified into a certain number of classifications from an assumed feature quantity estimation error and a variation of the characteristic quantity of the classification result, and the above-mentioned number of classifications is increased or decreased to obtain the above 2. The pulse signal classification device according to claim 1, further comprising means for classifying the pulse signals into a classification number that makes the evaluation appropriate, and means for correcting a feature quantity estimation error that is assumed from variations in the feature quantities of the classification results. 7. The pulse signal classification device according to claim 1, further comprising means for obtaining a pulse interval, and means for correcting a classification result based on the obtained pulse interval. 8. A means for correcting the classification of overlapping pulses that have received pulses from multiple signal sources simultaneously.
The pulse signal classification device described. 9. The pulse signal classification device according to claim 1, further comprising means for estimating a pulse repetition interval from a set of classified pulses and means for correcting the classification result using the estimated pulse repetition interval. 10. When classifying a pulse signal into several sets, a means for classifying the degree of belonging to each set by expressing it with an ambiguous value, a means for obtaining a pulse interval, and a correction method for classifying the degree of belonging. A pulse signal classification device comprising means for correcting the classification result based on the obtained pulse interval, which is used as a weighting value. 11. The pulse signal classification device according to claim 1, further comprising means for integrating a classification result. 12. The method according to claim 1, further comprising means for extracting discrete values of the pulse waveform, wherein the discrete value of the pulse waveform is used as a feature amount for performing the classification. Pulse signal classifier. 13. The pulse signal classification device according to claim 1, further comprising means for giving an initial value of classification when starting classification. 14. A means for accumulating a characteristic amount of a pulse signal of a known signal source, and a means for classifying a representative value of the pulse characteristic amount of each pulse train assuming that only the value of the accumulated characteristic amount is taken. The pulse signal classification device according to any one of claims 1 to 7.