JP3607846B2 - Clutter suppression device - Google Patents

Description

【００２３】
次に、第１のノッチフィルタ１および第２のノッチフィルタ２について説明する。
まず、送信パルス間隔が等間隔の場合について説明する。
想定されるクラッタ帯域幅に対応して、あらかじめ周波数０近傍を阻止域とする適当な高域通過ディジタルフィルタ（周波数０に零点を持つことが望ましい）を２種類用意しておく。それらの係数（実数）を第１のノッチフィルタ１用にａ_ｍ（ｍ＝０、１，・・・，Ｍ_１−１；Ｍ_１はインパルス応答長）、第２のノッチフィルタ２用にｂ_ｍ（ｍ＝０、１，・・・，Ｍ_２−１；Ｍ_２はインパルス応答長）とする。クラッタ中心周波数推定手段５で得られた推定値（平均した結果）は、上述のようにｆ_Ａ ^（ｂ）、ｆ_Ｂ ^（ｂ）である。第１のノッチフィルタ１の係数をα_ｍ ^（ｂ）、第２のノッチフィルタ２の係数β_ｍ ^（ｂ）とすると、式（３）のようになる。
【００２４】
【数２】

【００２５】
第１のノッチフィルタ１の出力信号ｙ_１（ｋ，ｎ）、第２のノッチフィルタ２の出力信号ｙ_２（ｋ，ｎ）はそれぞれ式（４）、（５）のようになる。レンジビン番号ｋがどのブロック番号ｂに属しているか注意する。
【００２６】
【数３】

【００２７】
次に、スタガトリガ方式の場合の第１のノッチフィルタ１および第２のノッチフィルタ２について説明する。
説明の前に、表記法を以下に記す。ｉ≦Ｌに対してはτ_ｉ＝ＰＲＩ_ｉ，ｉ＞Ｌに対してはτ_Ｌ＋１＝ＰＲＩ_１，τ_Ｌ＋２＝ＰＲＩ_２，・・・，τ_２Ｌ−１＝ＰＲＩ_Ｌ−１，τ_２Ｌ＝ＰＲＩ_Ｌ，τ_２Ｌ＋１＝ＰＲＩ_１，・・・とする。スタガトリガ方式の場合は、信号の表現において、ヒット変数ｎをパルス間隔に応じてｔ_ｎのように書くことにする。ただ、区別する必要がない場合は単にｎと記す。
【００２８】

【００２９】
文献３（Ｈ． Ｗ． Ｔｈｏｍａｓ， Ｎ． Ｐ． Ｌｕｔｔｅ， ａｎｄ Ｍ． Ｗ． Ｊｅｌｆｆｓ， “Ｄｅｓｉｇｎ ｏｆ ｍ．ｔ．ｉ． ｆｉｌｔｅｒｓ ｗｉｔｈ ｓｔａｇｇｅｒｅｄ ｐ．ｒ．ｆ： ａ ｐｏｌｅ−ｚｅｒｏ ａｐｐｒｏａｃｈ，” Ｐｒｏｃ． ＩＥＥ， ｖｏｌ． １２１， ｎｏ． １２， ｐ． １４６０−１４６６， １９７４）によれば、スタガトリガ方式の場合、従来の２重消去器などのようにフィルタ係数を時不変とするのではなく、処理を行おうとする信号のパルスの位置（あるいはパルス間隔）に応じてフィルタ係数を変えると、スタガトリガ方式でも周波数０に多重零点を持つ阻止域幅の広いノッチフィルタを得ることができる。本実施の形態でもそのような時変係数フィルタを使うことを想定して説明する（そのような時変係数フィルタしか使えないわけではない）。そのような時変係数ノッチフィルタのインパルス応答を、第１のノッチフィルタ１用にｈ_ｉ０，ｈ_ｉ１，・・・，ｈ_{ｉ 、 Ｍ１−１}、第２のノッチフィルタ２用にｇ_ｉ０，ｇ_ｉ１，・・・，ｇ_{ｉ 、 Ｍ２−１}とする。添字のｉ（ｉ＝０，１，・・・，Ｌ−１）は、パルス間隔に応じて係数が変わることを意味する。
【００３０】
クラッタ中心周波数推定手段５では、スタガトリガ方式でもそれを考慮せずに等間隔パルスとして信号を扱ってクラッタ中心周波数推定を行う。それで問題ないことを多数の例から確かめている。クラッタ中心周波数推定手段５で得られた推定値（平均した結果）は、上述のようにｆ_Ａ ^（ｂ）、ｆ_Ｂ ^（ｂ）である。第１のノッチフィルタ１の係数をＡ_ｉｍ ^（ｂ）、第２のノッチフィルタ２の係数Ｂ_ｉｍ ^（ｂ）とすると、式（６）、（７）のようになる。このように係数を設定すると、スタガトリガ方式でも深いノッチがそのまま周波数軸上で平行移動する。第１のノッチフィルタ１、第２のノッチフィルタ２の入出力信号の関係はそれぞれ式（８）、（９）のようになる。レンジビン番号ｋがどのブロック番号ｂに属しているか注意する。
【００３１】
【数４】

【００３２】
【数５】

【００３３】
次に、判定手段２０の動作について説明する。
これはスタガトリガ方式であってもなくても共通である。
最初に、全レンジビンのブロック分けを行う理由を説明する。
判定手段２０では、どの信号を選択するかを判定するが、その判定のための特徴量として、各信号の電力と、クラッタ中心周波数推定に利用するＡＲモデルの極を用いる。これらを用いてレンジビン毎に判定すると特徴量がばらついてしまい、判定結果がレンジビン毎に変化してしまう可能性がある。そうなると選択手段１０の出力信号の統計的性質もレンジビン毎に変わってしまう。これは、クラッタ抑圧処理に続くＣＦＡＲ（ｃｏｎｓｔａｎｔ ｆａｌｓｅ ａｌａｒｍ）による目標検出処理に悪影響を与える可能性がある。また、現実には、数十レンジビンでクラッタ中心周波数が大きく変わることはほとんどない。そこで、全レンジを適当なブロック長Ｂでブロック分けし、特徴量をレンジビンブロック内で平均して判定に用いる。判定はそのレンジビンブロック毎に行う。そうすれば、判定を行う回数は多くならないし、選択手段１０の出力信号の統計的性質もそのレンジビンブロック内では均一に保たれる。ＣＦＡＲ処理はそのレンジビンブロック内で行えばよい。
【００３４】
判定に用いる特徴量について説明する。
特徴量としては、受信信号、第１のノッチフィルタ１の出力信号、第２のノッチフィルタ２の出力信号の、各信号のレンジビンブロック内の平均電力、クラッタ中心周波数推定手段５で用いるＡＲモデルの極の絶対値をレンジビンブロック内で平均したものである。各信号の電力だけを比較したのでは、目標信号を消去してしまう可能性が大きいため、電力以外の特徴量であるＡＲモデルの極の絶対値の平均値を特徴量に加える。なお、各信号の電力を特徴量として利用せず、ＡＲモデルの極の絶対値の平均値だけを利用する判定手順も考えられる。レンジビンブロック番号ｂにおける平均電力は以下の式（１０）〜（１２）から算出する。Ｐ_０ ^（ｂ）、Ｐ_１ ^（ｂ）、Ｐ_２ ^（ｂ）はそれぞれ受信信号ｕ（ｋ，ｎ）、第１のノッチフィルタ１の出力信号ｙ_１（ｋ，ｎ）、第２のノッチフィルタ２の出力信号ｙ_２（ｋ，ｎ）のレンジビンブロック番号ｂにおける平均電力である。
【００３５】
【数６】

【００３６】
もう１つの特徴量であるＡＲモデルの極の絶対値の平均値について説明する。
図２は、文献２におけるクラッタ中心周波数推定法の概念図である。
図２は２次ＡＲモデルによる双峰性スペクトルを持つクラッタの中心周波数推定の場合で、１次ＡＲモデルによる単峰性スペクトルのクラッタ中心周波数推定の場合は、図２の「２次ＡＲモデリング」が１次ＡＲモデリングになり、各モデルに対して極と中心周波数推定値が１つのみで、図２の破線の矢印に関わる部分がない。
【００３７】
図２で、中心周波数推定に用いるヒット数をＮ_Ｂ、レンジビン数を注目セルｕ（ｋ，ｎ）の片側Ｋ_Ｂ、中心周波数推定に使わない注目セルの隣接レンジビン（ガードレンジ）数を片側Ｋ_Ｇとする。中心周波数推定には注目セルより過去のヒットを用いることにする。使用する信号の領域は、

である。
【００３８】
上のレンジビンの範囲の各ｋに対して、信号系列ｕ（ｋ，ｎ−Ｎ_Ｂ＋１），ｕ（ｋ，ｎ−Ｎ_Ｂ＋２），・・・，ｕ（ｋ，ｎ）をＢｕｒｇ法などで２次ＡＲモデリングし、２次ＡＲモデルの極を求める。２つの極をｑ_２１（ｋ，ｎ），ｑ_２２（ｋ，ｎ）とする。
【００３９】
次に、各レンジビンに対する極の実部と虚部それぞれに対してメディアン操作を行う。
これは目標周波数が推定値とならないようにするためのものである。メディアン操作で抽出された実部と虚部から偏角を求め、それから周波数を求める。この操作は極１、２の両方に対して行う。これで２つのクラッタ中心周波数推定値が求められる。この操作を数式で表現する。
【００４０】
各レンジビン（ｋ−（Ｋ_Ｂ＋Ｋ_Ｇ）〜ｋ−Ｋ_Ｇ−１、ｋ＋Ｋ_Ｇ＋１〜ｋ＋Ｋ_Ｂ＋Ｋ_Ｇの２Ｋ_Ｂレンジビン分）に対する２次ＡＲモデルの極が求まったとき、正規化されたクラッタ中心周波数推定値ｆ_ｃ２ｉ（ｋ，ｎ）（ｉ＝１，２）は式（１３）〜（１５）で求められる。式（１５）のｔａｎ^−１は、かっこ内の分母・分子の両方の符号を考慮して角度を求める。式（１３）、（１４）のＭｅｄはメディアン操作（中間値抽出）を意味する。メディアンをとる範囲は、上に示したレンジビン範囲である。式（１３）、（１４）はそれぞれ極の実部と虚部に相当する。以上の操作を、注目セルをずらしながら、全ヒット・全レンジビンに対して行う。同様に、単峰性スペクトルを仮定して、１次ＡＲモデリング後、メディアン操作によって極Ｑ_Ｒ１（ｋ，ｎ）＋ｊＱ_Ｉ１（ｋ，ｎ）とクラッタ中心周波数推定値ｆ_ｃ１（ｋ，ｎ）が求められる。
【００４１】
【数７】

【００４２】
判定手段２０における特徴量としてのＡＲモデルの極の絶対値のレンジビンブロック番号ｂ内の平均値（以後、混同のおそれがない限り、単にＡＲモデルの極の絶対値と呼ぶ）を、１次ＡＲモデルに対してＱ_１ ^（ｂ）、２次ＡＲモデルに対してＱ_２ｉ ^（ｂ）（ｉ＝１、２）とする。これは式（１６）、（１７）で求められる。
【００４３】
【数８】

【００４４】
ここで、ＡＲモデルの極の絶対値が特徴量となりうることを数値例で示す。そのため、計算機で疑似的な受信信号を生成した。等間隔パルスで、パルス繰り返し周波数ＰＲＦ＝１０００Ｈｚ、レンジビン数３００、ヒット数１２８である。ヒット数はスペクトルの表示の都合上、多く取った。固定クラッタは０Ｈｚを中心周波数とし、移動クラッタは−３２０Ｈｚが中心周波数である。ただし、移動クラッタは距離が遠くなるにつれて中心周波数がやや０に近づく。また、第２１０レンジビンにドップラー周波数５０Ｈｚの目標信号が存在する。クラッタの強度は、距離が遠くなるにつれて小さくなるようにした。特に固定クラッタの減衰が大きくなるようにした。固定クラッタの第２００レンジビン以遠は雑音レベルとほとんど変わらない。
【００４５】
図３は、生成した信号のスペクトルである。
第１、第５０、第１００、第１５０、第２１０レンジビンにおけるスペクトルを示す。凡例の＃１が第１レンジビンにおけるスペクトルで、以下同様である。固定クラッタのスペクトルレベルは距離が遠くなるにつれて減衰していくことがわかる。第２１０レンジビンのスペクトルで、５０Ｈｚ付近に目標信号によるピークが存在する。
【００４６】
図４は、レンジに対する式（１６）と（１７）のＡＲモデルの極の絶対値を示したものである。
レンジビンブロック長Ｂ＝２５である。ここで、１次ＡＲモデルによるもの（実線）と、２次ＡＲモデルによるものの一方（破線）は、比較的大きな値を保っているが、２次ＡＲモデルによるもののもう一方（一点鎖線）は、距離に従って減衰していくことがわかる。これは、固定クラッタの減衰に対応している。
【００４７】
このようにして色々と調べた結果、ＡＲモデルの極の絶対値について以下のことがわかった。
・クラッタが存在しないときは、１次ＡＲモデルの極の絶対値は小さくなる。
・単峰性スペクトルの場合は、１次ＡＲモデルの極の絶対値と、２次ＡＲモデルの極の絶対値の一方は比較的大きい値を取る。
・双峰性スペクトルの場合は、２次ＡＲモデルの２つの極の絶対値はともに比較的大きい値を取る。
【００４８】
これより、ＡＲモデルの極の絶対値は、スペクトルの単峰性や双峰性を判断する材料となることがわかった。文献２のクラッタ中心周波数推定法は、目標信号を保存させるために目標信号のスペクトルのピークを捕らえないようにしている。その過程で得られるＡＲモデルの極Ｑ_Ｒ１（ｋ，ｎ）＋ｊＱ_Ｉ１（ｋ，ｎ）、Ｑ_Ｒ２ｉ（ｋ，ｎ）＋ｊＱ_Ｉ２ｉ（ｋ，ｎ）（ｉ＝１、２）についても同様である。従って、式（１６）と（１７）のＡＲモデルの極の絶対値を特徴量として利用すると、目標信号によるスペクトルのピークを無視して、クラッタによるスペクトルのピークの数だけを数えられる。
【００４９】
判定手段２０によって選択すべき信号の決定手順について記す。
クラッタ中心周波数推定手段５におけるＡＲモデルの極の絶対値だけを用いる第１の手順と、それに加えて、式（１０）〜（１２）の、受信信号、第１のノッチフィルタ１の出力信号、第２のノッチフィルタ２の出力信号の、各信号のレンジビンブロック内の平均電力Ｐ_０ ^（ｂ）、Ｐ_１ ^（ｂ）、Ｐ_２ ^（ｂ）を特徴量としてともに用いる第２の手順を記す。スペクトルが単峰性か双峰性か、あるいはクラッタが存在しないかを判定するための、ＡＲモデルの極の絶対値のしきい値をＱ_ｔｈとする。
【００５０】
〈第１の判定手順〉
レンジビンブロック番号ｂ（ｂ＝１，２，・・・，Ｋ／Ｂ）について、
（１）式（１６）のＱ_１ ^（ｂ）がＱ_ｔｈ未満であれば、クラッタが存在しないと判断し、そのレンジビン番号ｂに属する範囲のレンジビンに対しては、図１のＡ、すなわち受信信号を選択する。
（２）式（１７）のＱ_２１ ^（ｂ）、Ｑ_２２ ^（ｂ）の一方がＱ_ｔｈ未満であれば、クラッタスペクトルは単峰性であると判断し、図１のＢ、すなわち、第１のノッチフィルタ１の出力信号を選択する。
（３）以上のいずれでもなければ、クラッタスペクトルは双峰性であると判断し、図１のＣ、すなわち、第２のノッチフィルタ２の出力信号を選択する。
【００５１】
〈第２の判定手順〉
図５は判定手段２０における第２の判定手順を示したフローチャートである。図５に従って手順を説明する。受信機雑音電力と関係して、２つの電力しきい値を設ける。小さい方のしきい値をＰ_{ｔｈ ｍｉｎ}、大きい方のしきい値をＰ_{ｔｈ ｍａｘ}とする。受信機雑音電力がこの２つのしきい値の間に入るようにする。
【００５２】
レンジビンブロック番号ｂ（ｂ＝１，２，・・・，Ｋ／Ｂ）について、
受信信号、第１のノッチフィルタ１の出力信号、第２のノッチフィルタ２の出力信号の各信号のレンジビンブロック内の平均電力である式（１０）〜（１２）のＰ_０ ^（ｂ）、Ｐ_１ ^（ｂ）、Ｐ_２ ^（ｂ）のうち、最小のものをＰ_ｍｉｎ ^（ｂ）とする（ステップＳ１）。電力の最も小さい信号（図１のＡ、Ｂ、Ｃのどれか）を選んで、ステップＳ２へ進む。
【００５３】
ステップＳ２：以下の３つの条件で分岐
ステップＳ１で選ばれた信号の電力Ｐ_ｍｉｎ ^（ｂ）がＰ_{ｔｈ ｍｉｎ}未満の場合（ステップＳ２．１）：
２番目に小さい信号電力がＰ_{ｔｈ ｍａｘ}より大きい場合は、ステップＳ１で選んだ信号を保持する。そうでない場合は、Ｐ_{ｔｈ ｍｉｎ}以上で最小の電力の信号を選ぶ（ステップＳ２．１．１）。そして、ステップＳ３へ進む。
ステップＳ１で選ばれた信号の電力がＰ_{ｔｈ ｍｉｎ}以上でＰ_{ｔｈ ｍａｘ}以下の場合（ステップＳ２．２）：
他の信号電力がこの範囲内にあれば、通過したノッチフィルタ次数が最も低い信号を選ぶ（ステップＳ２．２．１）。他の信号電力がこの範囲内になければステップＳ１で選んだ信号を保持する。そして、ステップＳ３へ進む。
ステップＳ１で選ばれた信号の電力がＰ_{ｔｈ ｍａｘ}より大きい場合（ステップＳ２．３）：
ステップＳ１で選んだ信号を最終選択結果として、選択操作を終了する。
【００５４】
次に、ステップＳ３において、１次ＡＲモデルの極の絶対値Ｑ_１ ^（ｂ）がＱ_ｔｈ未満であるなら、クラッタがないと判断して、受信信号（図１のＡ）を選択して（ステップＳ３．１）、選択操作を終了する。そうでない場合はステップＳ２までで選択した信号を保持して、ステップＳ４へ進む。
【００５５】
ステップＳ４において、ここまでで選択された信号が第１のノッチフィルタ１出力信号（図１のＢ）なら、その信号を選択結果として（ステップＳ４．１）、選択操作を終了する。そうでない場合はステップＳ５へ進む。
【００５６】
ステップＳ５において、１次ＡＲモデルの極の絶対値Ｑ_１ ^（ｂ）がＱ_ｔｈ以上かつ２次ＡＲモデルの２つの極の絶対値Ｑ_２１ ^（ｂ）、Ｑ_２２ ^（ｂ）の両方か一方がＱ_ｔｈ未満なら、クラッタスペクトルは単峰性と判断して第１のノッチフィルタ１の出力信号（図１のＢ）を選択結果とする（ステップＳ４．１）。そうでない場合は、第２のノッチフィルタ２の出力信号（図１のＣ）を選択結果とする（ステップＳ５．１）。以上で手順を終了する。
【００５７】
第２の手順における各ステップの意味を記す。
ステップＳ２．１は、既にクラッタが抑圧されているのに、さらにフィルタを通した信号を選ばないようにするものである。必要以上に電力が小さい信号はブラインド領域が広いと解釈する。
ステップＳ２．２は、電力しきい値範囲内であれば、雑音電力までクラッタが抑圧されたとみなし、通過するノッチフィルタの次数がなるべく低くなるように信号を選択するものである。
ステップＳ２．３は、どの信号を選んでもクラッタが消え残ってしまったが、なるべく消え残りが少ない信号を選択するものである。
【００５８】
ステップＳ３は、クラッタの有無をＡＲモデルの極の絶対値から判断するものである。
ステップＳ４は、クラッタスペクトルが単峰性か双峰性かを信号電力から判定するものである。
ステップＳ５は、さらに、クラッタスペクトルが単峰性か双峰性かをＡＲモデルの極から判定するものである。これは、本来単峰性スペクトルなのに誤って双峰性スペクトルと判断されているものを修正するために行うが、特に目標信号を保持するためのものである。電力だけで判断すると、目標信号を誤って抑圧してしまう可能性がある。
【００５９】
以上のように、実施の形態１によれば、判定手段２０で利用する特徴量が１コヒーレントプロセッシングインターバル内の受信信号で算出可能であるため、従来の技術のように数回のスキャンを必要としないで実際のクラッタの距離に関する分布に適応してクラッタ抑圧を行うことが可能となる。さらに、スタガトリガ方式でも深いノッチがそのまま周波数軸上で平行移動するようにノッチフィルタの係数を設定するため、移動クラッタ抑圧性能が高い。また、特徴量の算出には多数のヒット数を必要としないため、１コヒーレントプロセッシングインターバルあたりのパルスヒット数が少ない捜索レーダでも良好なクラッタ抑圧性能を得ることができる。
【００６０】
実施の形態２．
図６は、この発明の実施の形態２に係わるクラッタ抑圧装置の構成図である。この実施の形態２におけるクラッタ抑圧装置は、通常レーダ装置を構成する一要素であり、レーダ装置に組み込んだ形で使用する。ここで、実施の形態１（図１）との違いは、図１では第１のノッチフィルタ１に単峰性スペクトルのクラッタ抑圧と、双峰性スペクトルを持つクラッタの一方の抑圧を兼ねていた。しかし本実施の形態２においては、図６に示すように、第３のノッチフィルタ３（ＦＩＲ形）を設けて、これを単峰性スペクトルクラッタ抑圧専用とし、第１のノッチフィルタ１は双峰性スペクトルを持つクラッタの一方のクラッタ抑圧専用とするものである。この点を除けば、実施の形態１と動作は同じである。
【００６１】
以下、実施の形態２について、主に実施の形態１との相違点について説明する。
まず、送信パルス間隔が等間隔の場合について説明する。第１のノッチフィルタ１、第２のノッチフィルタ２、第３のノッチフィルタ３は、クラッタ中心周波数推定手段５での推定結果に基づいたノッチ周波数を持つ。第１のノッチフィルタ１、第２のノッチフィルタ２は式（２）に示されるｆ_０２１ ^（ｂ）、ｆ_０２２ ^（ｂ）、第３のノッチフィルタ３は式（１）で示されるｆ_０１ ^（ｂ）をノッチ周波数とする。想定されるクラッタ帯域幅に対応して、あらかじめ周波数０近傍を阻止域とする適当な高域通過ディジタルフィルタ（周波数０に零点を持つことが望ましい）を用意しておくことは実施の形態１と同様であるが、第３のノッチフィルタ３用にもう１つ用意する。場合によっては第１のノッチフィルタや第２のノッチフィルタ２で用いるものと同じでもよい。その係数（実数）を第３のノッチフィルタ３用にｃ_ｍ（ｍ＝０、１，・・・，Ｍ_３−１；Ｍ_３はインパルス応答長）とする。推定されたクラッタ中心周波数ｆ_０１ ^（ｂ）、ｆ_０２１ ^（ｂ）、ｆ_０２２ ^（ｂ）をノッチ周波数とする第１のノッチフィルタ１の係数をα_ｍ′^（ｂ）、第２のノッチフィルタ２の係数β_ｍ′^（ｂ）、第３のノッチフィルタ３の係数をγ_ｍ ^（ｂ）とすると、式（１８）のようになる。
【００６２】
【数９】

【００６３】
第１のノッチフィルタ１の出力信号ｙ_１（ｋ，ｎ）、第２のノッチフィルタ２の出力信号ｙ_２（ｋ，ｎ）、第３のノッチフィルタ３の出力信号ｙ_３（ｋ，ｎ）はそれぞれ式（１９）〜（２１）のようになる。レンジビン番号ｋがどのブロック番号ｂに属しているか注意する。
【００６４】
【数１０】

【００６５】
スタガトリガ方式の場合、処理を行おうとする信号のパルスの位置（あるいはパルス間隔）に応じてフィルタ係数を変えると、スタガトリガ方式でも周波数０に多重零点を持つ阻止域幅の広いノッチフィルタを得ることができる。本実施の形態でも、第３のノッチフィルタ３にそのような時変係数フィルタを用いるものとして説明する（そのような事変係数フィルタしか使えないわけではない）。そのような時変係数ノッチフィルタのインパルス応答を、第３のノッチフィルタ３用にｅ_ｉ０，ｅ_ｉ１，・・・，ｅ_{ｉ 、 Ｍ３−１}とする。添字のｉ（ｉ＝０，１，・・・，Ｌ−１）は、パルス間隔に応じて係数が変わることを意味する。第１のノッチフィルタ１の係数をＡ_ｉｍ′^（ｂ）、第２のノッチフィルタ２の係数Ｂ_ｉｍ′^（ｂ）、第３のノッチフィルタ３の係数Ｃ_ｉｍ ^（ｂ）とすると、それぞれ式（２２）〜（２４）のようになる。第１のノッチフィルタ１、第２のノッチフィルタ２、第３のノッチフィルタ３の入出力信号の関係はそれぞれ式（２５）〜（２７）のようになる。
【００６６】
【数１１】

【００６７】
【数１２】

【００６８】
判定手段２０における特徴量としての電力は、受信信号、第３のノッチフィルタ３の出力信号、第２のノッチフィルタ２の出力信号の、各信号のレンジビンブロック内の平均電力である。第１のノッチフィルタ１の出力信号は用いない。従って、式（１１）のＰ_１ ^（ｂ）の代わりに式（２８）で示すＰ_３ ^（ｂ）を用いる。第２の判定手順において、Ｐ_１ ^（ｂ）はＰ_３ ^（ｂ）に置き換える。選択肢とする信号は、図６のＡ、Ｂ、Ｃで示した受信信号ｕ（ｋ，ｎ）、第３のノッチフィルタ３の出力信号ｙ_３（ｋ，ｎ）、第２のノッチフィルタ２の出力信号ｙ_２（ｋ，ｎ）である。これらの点を除いて、判定手段２０の動作は実施の形態１と同じである。
【００６９】
【数１３】

【００７０】
以上のように、実施の形態２によれば、判定手段２０で利用する特徴量が１コヒーレントプロセッシングインターバル内の受信信号で算出可能であるため、従来の技術のように数回のスキャンを必要としないで実際のクラッタの距離に関する分布に適応してクラッタ抑圧を行うことが可能となる。加えて、スタガトリガ方式でも深いノッチがそのまま周波数軸上で平行移動するようにノッチフィルタの係数を設定するため、移動クラッタ抑圧性能が高い。また、特徴量の算出には多数のヒット数を必要としないため、１コヒーレントプロセッシングインターバルあたりのパルスヒット数が少ない捜索レーダでも良好なクラッタ抑圧性能を得ることができる。
【００７１】
実施の形態３．
実施の形態１と２ではクラッタが最大２つ存在するとしてきたが、それ以上（Ｉ個、Ｉ＞２）のクラッタに対処するためには、図１や図６の構成をそれぞれ拡張した図７あるいは図８の構成とする。
図７の構成について説明する。図７では、Ｉ個のノッチフィルタを縦続に接続する。それぞれのノッチフィルタは１つの周波数にノッチを持つ。クラッタ中心周波数推定手段５ではＩ次ＡＲモデルを使用してクラッタ中心周波数を推定する。推定されたＩ個の周波数をそれぞれＩ個のノッチフィルタのノッチ周波数として１個づつ割り当てる。各ノッチフィルタの係数は、第１および第２の実施形態と同様にして計算する。判定手段２０における判定手順としては、第１の判定手順をそのまま拡張できる。判定手段２０での判定結果に基づいて、選択手段１０では（Ｉ＋１）個の信号の中から１つを選択出力する。
【００７２】
次に、図８の構成について説明する。
図８では、１つのノッチフィルタ（３０−１）、２つのノッチフィルタを縦続に接続したもの（３０−２−１、３０−２−１），・・・，Ｉ個のノッチフィルタを縦続に接続したもの（３０−Ｉ−１，３０−Ｉ−２，・・・，３０−Ｉ−Ｉ）の、Ｉ組のノッチフィルタ群を並列接続する。合計Ｉ（Ｉ＋１）／２個のノッチフィルタを使用することになる。
【００７３】
個々のノッチフィルタは１つの周波数にノッチを持つ。クラッタ中心周波数推定手段５では、ｉ個（ｉ＝１，２，・・・，Ｉ）のノッチフィルタが縦続に接続されているノッチフィルタ群（第ｉノッチフィルタ群と呼ぶ）に対しては、ｉ個のクラッタが重畳していると仮定して、ｉ次ＡＲモデルを使ってクラッタ中心周波数推定を行い、その結果を第ｉノッチフィルタ群の各ノッチフィルタに１個づつ出力する。各ノッチフィルタの係数は、第１および第２の実施形態と同様にして計算する。これをｉ＝１，２，・・・，Ｉに対して行う。判定手段２０における判定手順としては、第１の判定手順をそのまま拡張できる。判定手段２０での判定結果に基づいて、選択手段１０では（Ｉ＋１）個の信号の中から１つを選択出力する。
【００７４】
これらの構成により、３つ以上のクラッタが存在しても、従来の技術のように数回のスキャンを必要としないで、１コヒーレントプロセッシングインターバル内の受信信号で実際のクラッタの距離に関する分布に適応してクラッタ抑圧を行うことが可能となる。加えて、スタガトリガ方式でも深いノッチがそのまま周波数軸上で平行移動するようにノッチフィルタの係数を設定するため、移動クラッタ抑圧性能が高い。また、特徴量の算出には多数のヒット数を必要としないため、１コヒーレントプロセッシングインターバルあたりのパルスヒット数が少ない捜索レーダでも良好なクラッタ抑圧性能を得ることができる。
【００７５】
次に、レーダ送信波形が等間隔パルスの場合、実施の形態２によってクラッタを抑圧できることを計算機シミュレーションにより示す。図３でスペクトルを示した信号を受信信号とする。第２１０レンジビンにドップラー周波数５０Ｈｚの目標が存在する。１レンジビンブロック長Ｂ＝２５レンジビンである。判定手段２０の判定手順は第２の手順を用いた。ＡＲモデルの極の絶対値のしきい値はＱ_ｔｈ＝０．３、電力のしきい値は、Ｐ_{ｔｈ ｍｉｎ}＝雑音電力−６ｄＢ、Ｐ_{ｔｈ ｍａｘ}＝雑音電力＋２ｄＢである。
【００７６】
選択手段１０の出力信号ｚ（ｋ，ｎ）のスペクトルを図９に示す。
これは、図３に対応して、第１、第５０、第１００、第１５０、第２１０レンジビンにおけるスペクトルである。目標信号は保存して、クラッタのみ抑圧できていることがわかる。選択手段１０で選択された出力信号は、第１５０レンジビンより近くは第２のノッチフィルタ２、それ以遠は第３のノッチフィルタ３であった。それぞれ、双峰性スペクトルのクラッタ、単峰性スペクトルのクラッタであると判定された。目標の存在する第２１０レンジビンでは単峰性スペクトルのクラッタと判断されており、その結果目標信号を消去せずにクラッタを抑圧できた。この処理は数回のアンテナスキャンを必要とせず、１コヒーレントプロセッシングインターバル内の受信信号で行っていることを強調しておく。
【００７７】
【発明の効果】
以上のように、この発明によれば、判定手段で利用する特徴量が１コヒーレントプロセッシングインターバル内の受信信号で算出可能であるため、数回のスキャンを必要としないで実際のクラッタの距離に関する分布に適応してクラッタ抑圧を行うことができる。
【００７８】
また、選択手段で選択出力すべき信号を複数の連続するレンジビンからなるブロック毎に判定するようにしたので、短時間にクラッタ抑圧を行うことができる。
【図面の簡単な説明】
【図１】この発明の実施の形態１に係るクラッタ抑圧装置の構成図である。
【図２】文献２によるクラッタ中心周波数推定法を模式的に表した図である。
【図３】計算機で疑似的に生成した固定クラッタと移動クラッタの両方と目標信号を含む信号のスペクトルである。
【図４】図３のスペクトルの信号に対して、レンジビンとＡＲモデルの極の絶対値の平均値の関係を示した図である。
【図５】判定手段２０における第２の判定手順を示したフローチャートである。
【図６】この発明の実施の形態２に係るクラッタ抑圧装置の構成図である。
【図７】この発明の実施の形態３に係るもので、３つ以上のクラッタに対処できるクラッタ抑圧装置の構成図である。
【図８】この発明の実施の形態３に係るもので、３つ以上のクラッタに対処できるクラッタ抑圧装置の他の構成図である。
【図９】この発明の実施の形態２に係るクラッタ抑圧処理の結果得られた信号の特定のレンジビンに対するスペクトルの説明図である。
【図１０】従来のクラッタ抑圧装置の概念的な構成図である。
【図１１】文献１で開示されたクラッタ抑圧装置の構成図である。
【符号の説明】
１ 第１のノッチフィルタ、２ 第２のノッチフィルタ、３ 第３のノッチフィルタ、４ 第Ｉのノッチフィルタ、５ クラッタ中心周波数推定手段、１０ 選択手段、２０ 判定手段、３０−１ ノッチフィルタ♯１、３０−２−１ ノッチフィルタ♯２−１、３０−２−２ ノッチフィルタ♯２−２、３０−Ｉ−１ノッチフィルタ♯Ｉ−１、３０−Ｉ−２ ノッチフィルタ♯Ｉ−２、３０−Ｉ−Ｉ ノッチフィルタ♯Ｉ−Ｉ。[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a clutter suppression device that suppresses unnecessary reflected waves (clutter) that are unintentionally received by a radar device.
[0002]
[Prior art]
In a radar apparatus, suppression of unnecessary reflected echo (clutter) from the ground, sea surface, clouds, rain, and the like is essential for target detection.
FIG. 10 is a diagram conceptually showing the configuration of the clutter suppression device disclosed in Japanese Patent Publication No. 3-2433.
In FIG. 10, 101 is a fixed clutter eraser for suppressing fixed clutter such as ground clutter and sea clutter, and 102 is an amplitude / phase corrector for correcting the amplitude and phase of the output signal of the fixed clutter eraser 101. , 103 are moving clutter erasers for suppressing clutter that the reflection source moves, such as weather clutter. In FIG. 10, in order to simplify the notation, the input signal is a complex signal in which the in-phase signal is a real part and the quadrature signal is an imaginary part.
[0003]
In the above configuration, the fixed clutter eraser 101 suppresses clutter from a reflection source that is very slow even if a reflection source such as a ground clutter or a sea clutter does not move or moves. However, if there is only a moving clutter such as a weather clutter, it cannot be erased only by the fixed clutter eraser 101. Therefore, a moving clutter eraser 103 is provided to suppress this. The amplitude / phase corrector 102 corrects the amplitude and phase of the output signal of the fixed clutter canceller 101 so that the moving clutter looks like a fixed clutter. The actual condition of the fixed clutter canceller 101 and the moving clutter canceller 103 is a high-pass filter having a notch frequency of zero. The transfer function is 1-z^-1Single eraser or (1-z^-1)²The double eraser is an example.
[0004]
In practice, fixed clutter does not extend to the search range of the search radar. Usually it extends only to a short distance. Moving clutter is more complex. Depending on the pulse interval in the radar, there may be secondary echoes. That is, the distribution regarding the actual clutter distance and the Doppler frequency is complicated. On the other hand, in the configuration shown in FIG. 10, both the fixed clutter eraser 101 and the moving clutter eraser 103 are applied over the entire distance. For example, if a moving clutter exists only at a long distance but a target signal having a Doppler frequency substantially the same as that of the moving clutter exists at a short distance, the target signal is deleted.
[0005]
On the other hand, a clutter suppression apparatus that performs clutter suppression in accordance with a distribution related to an actual clutter distance is disclosed in Reference 1 (A. Wojtkiewicz and M. Tuszynsky, “Polish radical technology Part V. Adaptive MTI filters”. Transactions on Aerospace and Electronic Systems, vol. 27, No. 5, September 1991).
[0006]
FIG. 11 is a block diagram showing the clutter suppression device disclosed in the above document.
11, 111 is a fixed clutter suppression filter for suppressing fixed clutter, 112 is a moving clutter suppression filter for suppressing moving clutter, and 113 is a clutter parameter estimator / clutter map generator for generating a clutter map. , 114 is a filter controller that individually controls on / off of the fixed clutter suppression filter and the moving clutter suppression filter based on the clutter map in the clutter parameter estimator / clutter map generator 113. In FIG. 11, in order to simplify the notation, the input signal is a complex signal in which the in-phase signal is a real part and the quadrature signal is an imaginary part.
[0007]
In the above configuration, the clutter parameter estimator / clutter map generator 113 estimates the received clutter power, the output power of the fixed clutter suppression filter 111, and the Doppler frequency of the moving clutter, and generates a clutter map. Unlike the configuration shown in FIG. 10, this depends on the distance, so that the presence or absence of a ground clutter (or sea clutter) or weather clutter can be known from the distance. Based on the generated clutter map, the filter controller 114 individually controls on / off of the fixed clutter suppression filter and the moving clutter suppression filter. Therefore, the clutter suppression can be performed in conformity with the distribution related to the actual clutter distance.
[0008]
[Problems to be solved by the invention]
However, the configuration shown in FIG. 11 described above has the following problems.
The presence / absence of the target signal cannot be determined from the power of the input / output signal of the fixed clutter suppression filter 111. To recognize the presence of the target signal in order to avoid erasure, the antenna beam must be scanned several times to update the clutter map. As disclosed in Document 1 (FIG. 26 and FIG. 28), the stop band attenuation of the moving clutter suppression filter cannot be sufficiently obtained particularly during stagger triggering. This means that the suppression capability of the mobile clutter is low.
[0009]
The present invention has been made to solve the above-described problems. In one coherent processing interval, clutter suppression is performed by adapting to the distribution related to the actual clutter distance in consideration of the presence of the target signal. It is possible to provide a clutter suppression device that can perform the above and has high moving clutter suppression performance.
[0010]
[Means for Solving the Problems]
The clutter suppression device according to the present invention is a clutter suppression device that suppresses clutter, which is an unnecessary reflected echo received by a radar, in which a clutter center frequency included in a received signal converted into a digital in-phase / quadrature signal is determined. Clutter center frequency estimation means that estimates based on the maximum entropy method on the assumption that there are a maximum of a plurality of I, a plurality of cascaded I notch filters that receive the reception signal, the reception signal, and the plurality of Based on the pole of an AR (auto-regressive) model obtained by the selection means for selectively outputting one of the I notch filter output signals as input and the clutter center frequency estimation means, The signal to be selected and output by the selection means is a block consisting of a plurality of continuous range bins. A plurality of I notch filters, wherein each of the plurality of I clutter center frequencies estimated by the clutter center frequency estimation unit is set as a zero point of a frequency characteristic. To do.
[0011]
In addition, the determining means determines a signal to be selectively output by the selecting means based on the pole of the AR model obtained by the clutter center frequency estimating means and the respective powers of the received signal and the plurality of notch filter output signals. The determination is made for each block composed of a plurality of continuous range bins.
[0012]
A clutter suppression device according to another invention is a clutter suppression device that suppresses clutter, which is an unnecessary reflected echo received by a radar. And i (i = 1, 2,..., I) cascade connection with clutter center frequency estimation means for estimating the number of clutters based on the maximum entropy method on the assumption that there are one to a plurality of I clutters A total of I (I + 1) / 2 notch filters in which the notch filters are connected in parallel with the received signal as input for i = 1, 2,..., I, the received signal, and the i number of filters. (I = 1, 2,..., I) Selection means for inputting a plurality of I output signals at the final stage of the cascade-connected notch filters and selecting and outputting one of them. Determination means for determining, based on the pole of the AR model obtained by the clutter center frequency estimation means, a signal to be selected and output by the selection means for each block composed of a plurality of continuous range bins. = 1, 2,..., I) The frequency that is the zero point of each of the cascade-connected notch filters is an estimated value by the clutter center frequency estimating means when it is assumed that there are i clutters. It is a feature.
[0013]
Further, the determination means includes a final stage of a notch filter cascade-connected to the i poles (i = 1, 2,..., I) of the AR model pole obtained by the clutter center frequency estimation means and the received signal. On the basis of the power of each of the plurality of I output signals, a signal to be selectively output by the selection means is determined for each block composed of a plurality of continuous range bins.
[0014]
DETAILED DESCRIPTION OF THE INVENTION
Embodiment 1 FIG.
1 is a block diagram showing a clutter suppression apparatus according to Embodiment 1 of the present invention. The clutter suppression device in this embodiment is one element that normally constitutes a radar device, and is used in a form incorporated in the radar device.
In FIG. 1, reference numeral 1 denotes a first notch filter (FIR type, FIR: Finite Impulse Response) for suppressing one of two superimposed clutters having a monomodal spectrum or two clutters having a bimodal spectrum. ) 2 is a second notch filter (FIR type) for suppressing the clutter having a bimodal spectrum that remains undisappeared in the first notch filter 1, and 5 is a clutter center for estimating the clutter center frequency in the received signal. Frequency estimation means 10 is a selection means for selecting and outputting one of three signals of a received signal, an output signal of the first notch filter 1 and an output signal of the second notch filter 2, and 20 is a selection means It is a determination means for determining which of the three input signals is selected.
[0015]
Here, the bimodal spectrum refers to a spectrum having two peaks by overlapping a fixed clutter and a moving clutter. In addition, the unimodal spectrum is a spectrum in which only one of a fixed clutter and a moving clutter exists as a clutter and has one peak. The spectrum peak due to the target signal is not counted.
[0016]
The operation of the first embodiment will be described below with reference to FIG.
As a radar transmission signal, a pulse signal or a pulse signal that is frequency-modulated or code-modulated to perform pulse compression is assumed. The signal is expressed in a two-dimensional format of range bin and hit. The received signal is u (k, n), and the output signal of the first notch filter 1 is y₁(K, n), the output signal of the second notch filter 2 is y₂Let (k, n). k is a range bin number (k = 1, 2,..., K), and n is a hit number (n = 1, 2,..., N). K and N are the number of range bins and the number of hits, respectively, during one coherent processing interval. The received signal is converted into a digital in-phase / quadrature signal by a receiver, an A / D converter or the like. To simplify the notation, each signal is a complex digital signal with the in-phase signal as a real part and the quadrature signal as an imaginary part.
[0017]
The first notch filter 1 suppresses a fixed clutter or a moving clutter out of a clutter of a unimodal spectrum or a clutter of a bimodal spectrum. The second notch filter 2 suppresses the clutter remaining unerased by the first notch filter 1 out of the bimodal spectrum clutter. Both the notch frequencies of the first notch filter 1 and the second notch filter 2 have a notch at the frequency estimated by the clutter center frequency estimating means 5. For this purpose, first, the clutter center frequency estimation means 5 assumes that the received signal is a clutter of a bimodal spectrum, and uses a second-order AR (auto-regressive) model to determine the clutter center frequency (the center frequency of each of the two clutters). ). This is described in Reference 2 (Harazawa, Mano: “Adaptive MTI using median filter”, IEICE Transactions B-II, vol. J79-B-II, No. 12, pp. 1013-1021, Nov., 1996) is used. Based on the maximum entropy method, estimation using an AR model is performed.
[0018]
The clutter center frequency is estimated for each range bin. Since a bimodal spectrum is assumed, there are two types of center frequency estimation results. Estimated value f of one of them_A ^(B)The clutter suppression processing is performed by the first notch filter 1 having a notch frequency. The output signal of the first notch filter 1 is output to the determination means 20 and the selection means 10 and also output to the second notch filter 2. In the second notch filter 2, the other one of the clutter center frequency estimation values just above f._B ^(B)Is subjected to clutter suppression processing by the second notch filter 2 having the notch frequency as a notch frequency, and the result is output to the determination means 20 and the selection means 10.
[0019]
In the determination means 20, the received signal power, the output signal power of the first notch filter 1, the output signal power of the second notch filter 2, and the poles of the AR model used for clutter center frequency estimation in the clutter center frequency estimation means 5. Are used together to determine which of the three signals should be output. In the determination by the determination means 20, there is a method in which the received signal power, the first notch filter 1 output signal power, and the second notch filter 2 output signal power are not used. Based on the determination result of the determination unit 20, the selection unit 10 selectively outputs one of the three signals. The output signal of the selection means 10 is subsequently used for target detection and the like. The detailed operation of the determination unit 20 will be described later. F_A ^(B), F_B ^(B)(B) is a block number when the total range bin number K is divided into blocks by an appropriate block length B as will be described later. b = 1, 2,..., K / B. f_A ^(B), F_B ^(B)Is the pulse repetition frequency PRF (in the case of the stagger trigger method, the average pulse interval PRI_avThe frequency normalized by the reciprocal of Hereinafter, the frequency is a normalized frequency unless otherwise specified.
[0020]
f_A ^(B)And f_B ^(B)The calculation of will be described.
In the estimation method of Document 2, the clutter center frequency estimation is performed for each range bin and hit of the received signal, so the clutter center frequency estimation value depends on the range bin k and the hit n. Assuming bimodal spectral clutter, a second-order AR model is used. Since there are two clutter center frequency estimates,_c21(K, n), f_c22Let (k, n). In addition, the clutter center frequency estimation using the first-order AR model is performed at the same time assuming the clutter of the unimodal spectrum. The clutter center frequency estimate assuming a unimodal spectrum is f_c1Let (k, n). Basically, of the two center frequency estimation values obtained by assuming the bimodal spectrum clutter, the one closer to the clutter center frequency estimation value assuming the unimodal spectrum is used for the first notch filter 1. The other is used for the second notch filter 2. f_A ^(B), F_B ^(B)Uses the hit and the average within its range bin block. Note that the averaging operation is performed on a complex number having an absolute value of 1 whose frequency is a declination, and the resulting declination is used as an average frequency.
[0021]
f_c1(K, n), f_c21(K, n), f_c22The average of (k, n) within hit and range bin block number b is f₀₁ ^(B), F₀₂₁ ^(B), F₀₂₂ ^(B)And shown in equations (1) and (2). f₀₂₁ ^(B), F₀₂₂ ^(B)F₀₁ ^(B)F closer to_A ^(B), The other is f_B ^(B)And
[0022]
[Expression 1]

[0023]
Next, the first notch filter 1 and the second notch filter 2 will be described.
First, the case where transmission pulse intervals are equal will be described.
Two types of suitable high-pass digital filters (desirably having a zero at frequency 0) are prepared in advance corresponding to the assumed clutter bandwidth. Those coefficients (real numbers) are a for the first notch filter 1._m(M = 0, 1, ..., M₁-1; M₁Is the impulse response length), b for the second notch filter 2_m(M = 0, 1, ..., M₂-1; M₂Is the impulse response length). The estimated value (averaged result) obtained by the clutter center frequency estimating means 5 is f as described above._A ^(B), F_B ^(B)It is. The coefficient of the first notch filter 1 is α_m ^(B), Coefficient β of the second notch filter 2_m ^(B)Then, it becomes like Formula (3).
[0024]
[Expression 2]

[0025]
Output signal y of the first notch filter 1₁(K, n), the output signal y of the second notch filter 2₂(K, n) is represented by equations (4) and (5), respectively. Note which block number b the range bin number k belongs to.
[0026]
[Equation 3]

[0027]
Next, the first notch filter 1 and the second notch filter 2 in the case of the stagger trigger method will be described.
Before the explanation, the notation is described below. τ for i ≦ L_i= PRI_i, I> L, τ_{L + 1}= PRI₁, Τ_{L + 2}= PRI₂, ..., τ_2L-1= PRI_L-1, Τ_2L= PRI_L, Τ_{2L + 1}= PRI₁, .... In the case of the stagger trigger method, in the signal representation, the hit variable n is set to t according to the pulse interval._nI will write like this. However, when it is not necessary to distinguish, it is simply written as n.
[0028]

[0029]
Reference 3 (H. W. Thomas, N. P. Lutete, and M. W. Jelffs, “Design of m. T. Filters with staggered p. R. F: a pole-zero approach, E. Pro. , Vol. 121, no. 12, p. 1460-1466, 1974), in the case of the stagger trigger method, the filter coefficient is not time-invariant as in the conventional double eraser, but the processing is performed. If the filter coefficient is changed according to the pulse position (or pulse interval) of the signal to be obtained, a notch filter having a wide stop band width having multiple zeros at frequency 0 can be obtained even by the stagger trigger method. This embodiment will also be described assuming that such a time-varying coefficient filter is used (not only such a time-varying coefficient filter can be used). The impulse response of such a time-varying coefficient notch filter is expressed as h for the first notch filter 1._i0, H_i1, ..., h_{i , M1-1}G for the second notch filter 2_i0, G_i1, ..., g_{i , M2-1}And The subscript i (i = 0, 1,..., L−1) means that the coefficient changes according to the pulse interval.
[0030]
The clutter center frequency estimation means 5 estimates the clutter center frequency by treating the signals as equally spaced pulses without considering the stagger trigger method. We have confirmed from many examples that there is no problem. The estimated value (averaged result) obtained by the clutter center frequency estimating means 5 is f as described above._A ^(B), F_B ^(B)It is. The coefficient of the first notch filter 1 is A_im ^(B), Coefficient B of the second notch filter 2_im ^(B)Then, equations (6) and (7) are obtained. When the coefficient is set in this way, the deep notch is translated on the frequency axis as it is even in the stagger trigger method. The relationship between the input / output signals of the first notch filter 1 and the second notch filter 2 is expressed by equations (8) and (9), respectively. Note which block number b the range bin number k belongs to.
[0031]
[Expression 4]

[0032]
[Equation 5]

[0033]
Next, the operation of the determination unit 20 will be described.
This is common regardless of the stagger trigger method.
First, the reason for performing block division of all range bins will be described.
The determination means 20 determines which signal is selected, and uses the power of each signal and the pole of the AR model used for clutter center frequency estimation as the feature quantity for the determination. If these are used for determination for each range bin, the feature amount varies, and the determination result may vary for each range bin. When this happens, the statistical properties of the output signal of the selection means 10 will also change for each range bin. This may adversely affect target detection processing by CFAR (constant false alarm) following clutter suppression processing. In reality, the clutter center frequency hardly changes greatly in several tens of range bins. Therefore, the entire range is divided into blocks with an appropriate block length B, and the feature values are averaged in the range bin block and used for determination. The determination is made for each range bin block. By doing so, the number of determinations does not increase, and the statistical properties of the output signal of the selection means 10 are also kept uniform in the range bin block. The CFAR process may be performed within the range bin block.
[0034]
A feature amount used for determination will be described.
As the feature amount, the received signal, the output signal of the first notch filter 1, the output signal of the second notch filter 2, the average power in the range bin block of each signal, the AR model used in the clutter center frequency estimation means 5 The absolute value of the pole is averaged within the range bin block. If only the power of each signal is compared, there is a high possibility that the target signal will be deleted. Therefore, the average absolute value of the poles of the AR model, which is a feature quantity other than power, is added to the feature quantity. Note that a determination procedure that uses only the average of the absolute values of the poles of the AR model without using the power of each signal as a feature amount is also conceivable. The average power in the range bin block number b is calculated from the following formulas (10) to (12). P₀ ^(B), P₁ ^(B), P₂ ^(B)Are the received signal u (k, n) and the output signal y of the first notch filter 1, respectively.₁(K, n), the output signal y of the second notch filter 2₂This is the average power in the range bin block number b of (k, n).
[0035]
[Formula 6]

[0036]
The average value of the absolute values of the poles of the AR model, which is another feature amount, will be described.
FIG. 2 is a conceptual diagram of the clutter center frequency estimation method in Document 2.
FIG. 2 shows the case of estimating the center frequency of a clutter having a bimodal spectrum by a second order AR model. In the case of estimating the clutter center frequency of a single peak spectrum by a first order AR model, “Second order AR modeling” of FIG. Is the first-order AR modeling, and there is only one pole and center frequency estimation value for each model, and there is no portion related to the dashed arrow in FIG.
[0037]
In FIG. 2, the number of hits used for center frequency estimation is represented by N_B, The number of range bins on one side K of the cell of interest u (k, n)_B, The number of adjacent range bins (guard ranges) of the cell of interest not used for center frequency estimation_GAnd For the center frequency estimation, hits past from the cell of interest are used. The area of the signal used is

It is.
[0038]
For each k in the range of the upper range bin, the signal sequence u (k, n−N_B+1), u (k, n−N_B+2),..., U (k, n) are second-order AR modeled by the Burg method or the like to determine the poles of the second-order AR model. Q the two poles₂₁(K, n), q₂₂Let (k, n).
[0039]
Next, a median operation is performed on each of the real part and imaginary part of the pole for each range bin.
This is to prevent the target frequency from becoming an estimated value. The declination is obtained from the real part and the imaginary part extracted by the median operation, and then the frequency is obtained. This operation is performed for both poles 1 and 2. Thus, two clutter center frequency estimation values are obtained. This operation is expressed by a mathematical expression.
[0040]
Each range bin (k- (K_B+ K_G) ~ K-K_G-1, k + K_G+1 to k + K_B+ K_G2K_BNormalized clutter center frequency estimate f when the poles of the second order AR model for the range bin) are found_c2i(K, n) (i = 1, 2) is obtained by equations (13) to (15). Tan in formula (15)^-1Finds the angle in consideration of both the denominator and numerator signs in parentheses. Med in the equations (13) and (14) means median operation (intermediate value extraction). The range where the median is taken is the range bin range shown above. Expressions (13) and (14) correspond to the real part and imaginary part of the pole, respectively. The above operation is performed for all hits and all range bins while shifting the target cell. Similarly, assuming a unimodal spectrum, after first-order AR modeling, the pole Q_R1(K, n) + jQ_I1(K, n) and clutter center frequency estimate f_c1(K, n) is determined.
[0041]
[Expression 7]

[0042]
The average value in the range bin block number b of the absolute value of the pole of the AR model as the feature value in the determination means 20 (hereinafter simply referred to as the absolute value of the pole of the AR model unless there is a possibility of confusion) Q for model₁ ^(B)Q for secondary AR model_2i ^(B)(I = 1, 2). This is obtained by equations (16) and (17).
[0043]
[Equation 8]

[0044]
Here, a numerical example shows that the absolute value of the pole of the AR model can be a feature amount. Therefore, a pseudo received signal was generated by a computer. Pulses with equal intervals, pulse repetition frequency PRF = 1000 Hz, range bin number 300, hit number 128. A large number of hits was taken for convenience of spectrum display. The fixed clutter has a center frequency of 0 Hz, and the moving clutter has a center frequency of −320 Hz. However, the center frequency of the moving clutter approaches 0 as the distance increases. A target signal having a Doppler frequency of 50 Hz exists in the 210th range bin. The intensity of the clutter was made smaller with increasing distance. In particular, the attenuation of fixed clutter was increased. The fixed clutter beyond the 200th range bin is almost the same as the noise level.
[0045]
FIG. 3 shows the spectrum of the generated signal.
The spectrum in a 1st, 50th, 100th, 150th, 210th range bin is shown. Legend # 1 is the spectrum in the first range bin, and so on. It can be seen that the spectral level of the fixed clutter attenuates with increasing distance. In the spectrum of the 210th range bin, a peak due to the target signal exists in the vicinity of 50 Hz.
[0046]
FIG. 4 shows the absolute values of the poles of the AR model of equations (16) and (17) with respect to the range.
Range bin block length B = 25. Here, one of the primary AR model (solid line) and one of the secondary AR model (broken line) maintain a relatively large value, while the other of the secondary AR model (dashed line) is It turns out that it attenuates according to distance. This corresponds to the attenuation of the fixed clutter.
[0047]
As a result of various investigations as described above, it was found that the absolute values of the poles of the AR model were as follows.
• When there is no clutter, the absolute value of the pole of the primary AR model is small.
In the case of a unimodal spectrum, one of the absolute value of the pole of the primary AR model and the absolute value of the pole of the secondary AR model takes a relatively large value.
In the case of a bimodal spectrum, the absolute values of the two poles of the second-order AR model both take a relatively large value.
[0048]
From this, it was found that the absolute value of the pole of the AR model is a material for judging the monomodality and bimodality of the spectrum. The clutter center frequency estimation method of Document 2 is designed not to capture the spectrum peak of the target signal in order to preserve the target signal. The pole Q of the AR model obtained in the process_R1(K, n) + jQ_I1(K, n), Q_R2i(K, n) + jQ_I2iThe same applies to (k, n) (i = 1, 2). Therefore, if the absolute values of the poles of the AR model in the equations (16) and (17) are used as feature amounts, only the number of spectrum peaks due to clutter can be counted ignoring the spectrum peak due to the target signal.
[0049]
A procedure for determining a signal to be selected by the determination means 20 will be described.
The first procedure using only the absolute value of the pole of the AR model in the clutter center frequency estimation means 5, and in addition, the received signal, the output signal of the first notch filter 1, of the equations (10) to (12), The average power P in the range bin block of each signal of the output signal of the second notch filter 2₀ ^(B), P₁ ^(B), P₂ ^(B)A second procedure using both as features will be described. The absolute value threshold of the pole of the AR model to determine whether the spectrum is unimodal or bimodal, or no clutter is present, Q_thAnd
[0050]
<First determination procedure>
For the range bin block number b (b = 1, 2,..., K / B)
(1) Q in formula (16)₁ ^(B)Is Q_thIf it is less than that, it is determined that there is no clutter, and A in FIG. 1, that is, the received signal is selected for the range bin in the range belonging to the range bin number b.
(2) Q in formula (17)₂₁ ^(B), Q₂₂ ^(B)One of the Q is_thIf it is less than that, it is determined that the clutter spectrum is unimodal, and B in FIG. 1, that is, the output signal of the first notch filter 1 is selected.
(3) If none of the above, it is determined that the clutter spectrum is bimodal, and C in FIG. 1, that is, the output signal of the second notch filter 2 is selected.
[0051]
<Second determination procedure>
FIG. 5 is a flowchart showing a second determination procedure in the determination means 20. The procedure will be described with reference to FIG. In relation to the receiver noise power, two power thresholds are provided. Set the smaller threshold to P_{th min}, Let P be the larger threshold_{th max}And The receiver noise power is allowed to fall between these two thresholds.
[0052]
For range bin block number b (b = 1, 2,..., K / B)
P in the equations (10) to (12), which is the average power in the range bin block of each of the received signal, the output signal of the first notch filter 1 and the output signal of the second notch filter 2₀ ^(B), P₁ ^(B), P₂ ^(B)Of these, the smallest one is P_min ^(B)(Step S1). A signal having the smallest power (any one of A, B, and C in FIG. 1) is selected, and the process proceeds to step S2.
[0053]
Step S2: Branch under the following three conditions
The power P of the signal selected in step S1_min ^(B)Is P_{th min}If less than (step S2.1):
The second smallest signal power is P_{th max}If larger, the signal selected in step S1 is held. Otherwise, P_{th min}The signal with the minimum power is selected as described above (step S2.1.1). Then, the process proceeds to step S3.
The power of the signal selected in step S1 is P_{th min}P_{th max}In the following case (step S2.2):
If the other signal power is within this range, the signal having the lowest notch filter order passed is selected (step S2.2.1). If the other signal power is not within this range, the signal selected in step S1 is held. Then, the process proceeds to step S3.
The power of the signal selected in step S1 is P_{th max}If larger (step S2.3):
The selection operation is terminated with the signal selected in step S1 as the final selection result.
[0054]
Next, in step S3, the absolute value Q of the poles of the primary AR model₁ ^(B)Is Q_thIf not, it is determined that there is no clutter, the received signal (A in FIG. 1) is selected (step S3.1), and the selection operation is terminated. Otherwise, the signal selected up to step S2 is held and the process proceeds to step S4.
[0055]
In step S4, if the signal selected so far is the first notch filter 1 output signal (B in FIG. 1), that signal is used as a selection result (step S4.1), and the selection operation is terminated. Otherwise, the process proceeds to step S5.
[0056]
In step S5, the absolute value Q of the poles of the primary AR model₁ ^(B)Is Q_thThe absolute value Q of the two poles of the second order AR model₂₁ ^(B), Q₂₂ ^(B)Q or both_thIf it is less than that, the clutter spectrum is determined to be unimodal, and the output signal (B in FIG. 1) of the first notch filter 1 is selected (step S4.1). If not, the output signal (C in FIG. 1) of the second notch filter 2 is selected (step S5.1). This is the end of the procedure.
[0057]
The meaning of each step in the second procedure will be described.
In step S2.1, the signal that has passed through the filter is not selected even though the clutter is already suppressed. A signal with lower power than necessary is interpreted as having a large blind area.
In step S2.2, if within the power threshold range, it is considered that the clutter is suppressed to the noise power, and the signal is selected so that the order of the notch filter passing therethrough is as low as possible.
In step S2.3, the clutter disappears regardless of which signal is selected, but selects a signal with as little disappearance as possible.
[0058]
Step S3 determines the presence or absence of clutter from the absolute value of the pole of the AR model.
Step S4 determines whether the clutter spectrum is unimodal or bimodal from the signal power.
Step S5 further determines whether the clutter spectrum is unimodal or bimodal from the poles of the AR model. This is performed in order to correct what was originally determined to be a bimodal spectrum although it was originally a unimodal spectrum, and is particularly for holding a target signal. If it is determined only by electric power, the target signal may be erroneously suppressed.
[0059]
As described above, according to the first embodiment, since the feature amount used by the determination unit 20 can be calculated from the received signal within one coherent processing interval, several scans are required as in the conventional technique. It is possible to suppress clutter by adapting to the distribution related to the actual distance of the clutter. Further, even in the stagger trigger method, the coefficient of the notch filter is set so that the deep notch moves in parallel on the frequency axis, so that the moving clutter suppression performance is high. In addition, since a large number of hits is not required for calculating the feature amount, even a search radar having a small number of pulse hits per coherent processing interval can obtain good clutter suppression performance.
[0060]
Embodiment 2. FIG.
FIG. 6 is a block diagram of a clutter suppression device according to Embodiment 2 of the present invention. The clutter suppression device according to the second embodiment is a component that normally constitutes a radar device, and is used in a form incorporated in the radar device. Here, the difference from the first embodiment (FIG. 1) is that, in FIG. 1, the first notch filter 1 serves both as a clutter suppression of a single peak spectrum and a suppression of one of a clutter having a bimodal spectrum. . However, in the second embodiment, as shown in FIG. 6, a third notch filter 3 (FIR type) is provided, which is dedicated to unimodal spectrum clutter suppression, and the first notch filter 1 is a bimodal. One of the clutters having a sex spectrum is exclusively used for clutter suppression. Except this point, the operation is the same as in the first embodiment.
[0061]
Hereinafter, the difference between the second embodiment and the first embodiment will be mainly described.
First, the case where transmission pulse intervals are equal will be described. The first notch filter 1, the second notch filter 2, and the third notch filter 3 have notch frequencies based on the estimation result of the clutter center frequency estimation means 5. The first notch filter 1 and the second notch filter 2 are expressed as f shown in Equation (2).₀₂₁ ^(B), F₀₂₂ ^(B), The third notch filter 3 is expressed by the equation (1) f₀₁ ^(B)Is the notch frequency. Corresponding to the assumed clutter bandwidth, preparing an appropriate high-pass digital filter (desirably having a zero at frequency 0) in advance with a frequency near 0 is the same as in the first embodiment. Similarly, another one is prepared for the third notch filter 3. In some cases, it may be the same as that used in the first notch filter or the second notch filter 2. The coefficient (real number) is c for the third notch filter 3._m(M = 0, 1, ..., M₃-1; M₃Is the impulse response length). Estimated clutter center frequency f₀₁ ^(B), F₀₂₁ ^(B), F₀₂₂ ^(B)Is the coefficient of the first notch filter 1 with the notch frequency α_m′^(B), Coefficient β of the second notch filter 2_m′^(B), The coefficient of the third notch filter 3 is γ_m ^(B)Then, the equation (18) is obtained.
[0062]
[Equation 9]

[0063]
Output signal y of the first notch filter 1₁(K, n), the output signal y of the second notch filter 2₂(K, n), the output signal y of the third notch filter 3₃(K, n) is represented by equations (19) to (21), respectively. Note which block number b the range bin number k belongs to.
[0064]
[Expression 10]

[0065]
In the case of the stagger trigger method, if the filter coefficient is changed according to the pulse position (or pulse interval) of the signal to be processed, a notch filter having a wide stop band width having multiple zeros at the frequency 0 can be obtained even in the stagger trigger method. it can. Also in this embodiment, description will be made assuming that such a time-varying coefficient filter is used for the third notch filter 3 (only such an event variable coefficient filter can be used). The impulse response of such a time-varying notch filter is defined as e for the third notch filter 3._i0, E_i1, ..., e_{i , M3-1}And The subscript i (i = 0, 1,..., L−1) means that the coefficient changes according to the pulse interval. The coefficient of the first notch filter 1 is A_im′^(B), Coefficient B of the second notch filter 2_im′^(B), Coefficient C of the third notch filter 3_im ^(B)Then, equations (22) to (24) are obtained, respectively. The relationship between the input / output signals of the first notch filter 1, the second notch filter 2, and the third notch filter 3 is expressed by equations (25) to (27), respectively.
[0066]
[Expression 11]

[0067]
[Expression 12]

[0068]
The power as the characteristic amount in the determination unit 20 is the average power in the range bin block of each signal of the received signal, the output signal of the third notch filter 3, and the output signal of the second notch filter 2. The output signal of the first notch filter 1 is not used. Therefore, P in equation (11)₁ ^(B)Instead of P in the formula (28)₃ ^(B)Is used. In the second determination procedure, P₁ ^(B)Is P₃ ^(B)Replace with The signals to be selected are the received signal u (k, n) indicated by A, B, and C in FIG. 6 and the output signal y of the third notch filter 3.₃(K, n), the output signal y of the second notch filter 2₂(K, n). Except for these points, the operation of the determination means 20 is the same as that of the first embodiment.
[0069]
[Formula 13]

[0070]
As described above, according to the second embodiment, since the feature amount used by the determination unit 20 can be calculated from the received signal within one coherent processing interval, several scans are required as in the conventional technique. It is possible to suppress clutter by adapting to the distribution related to the actual distance of the clutter. In addition, even in the staggered trigger method, the coefficient of the notch filter is set so that the deep notch moves in parallel on the frequency axis, so that the moving clutter suppression performance is high. In addition, since a large number of hits is not required for calculating the feature amount, even a search radar having a small number of pulse hits per coherent processing interval can obtain good clutter suppression performance.
[0071]
Embodiment 3 FIG.
In Embodiments 1 and 2, it is assumed that there are a maximum of two clutters, but in order to cope with more (I, I> 2) clutters, the configurations of FIGS. 1 and 6 are expanded respectively. Or it is set as the structure of FIG.
The configuration of FIG. 7 will be described. In FIG. 7, I notch filters are connected in cascade. Each notch filter has a notch at one frequency. The clutter center frequency estimation means 5 estimates the clutter center frequency using an I-order AR model. The estimated I frequencies are assigned one by one as the notch frequencies of the I notch filters. The coefficient of each notch filter is calculated in the same manner as in the first and second embodiments. As a determination procedure in the determination means 20, the first determination procedure can be expanded as it is. Based on the determination result of the determination unit 20, the selection unit 10 selects and outputs one of (I + 1) signals.
[0072]
Next, the configuration of FIG. 8 will be described.
In FIG. 8, one notch filter (30-1), two notch filters connected in cascade (30-2-1, 30-2-1), ..., I notch filters in cascade The I notch filter groups of the connected ones (30-I-1, 30-I-2,..., 30-II) are connected in parallel. A total of I (I + 1) / 2 notch filters will be used.
[0073]
Each notch filter has a notch at one frequency. In the clutter center frequency estimating means 5, for a notch filter group (referred to as an i-th notch filter group) in which i (i = 1, 2,..., I) notch filters are connected in cascade, Assuming that i clutters are superimposed, clutter center frequency estimation is performed using an i-th AR model, and the result is output to each notch filter of the i-th notch filter group. The coefficient of each notch filter is calculated in the same manner as in the first and second embodiments. This is performed for i = 1, 2,. As a determination procedure in the determination means 20, the first determination procedure can be expanded as it is. Based on the determination result of the determination unit 20, the selection unit 10 selects and outputs one of (I + 1) signals.
[0074]
With these configurations, even if there are more than two clutters, the received signal within one coherent processing interval can be adapted to the distribution related to the actual clutter distance without requiring several scans as in the prior art. Thus, clutter suppression can be performed. In addition, even in the stagger trigger method, the coefficient of the notch filter is set so that the deep notch translates as it is on the frequency axis, so that the moving clutter suppression performance is high. In addition, since a large number of hits is not required for calculating the feature amount, even a search radar having a small number of pulse hits per coherent processing interval can obtain good clutter suppression performance.
[0075]
Next, it will be shown by computer simulation that clutter can be suppressed by the second embodiment when the radar transmission waveform is an equidistant pulse. A signal whose spectrum is shown in FIG. A target with a Doppler frequency of 50 Hz exists in the 210th range bin. One range bin block length B = 25 range bins. The determination procedure of the determination means 20 used the second procedure. The absolute value threshold of the pole of the AR model is Q_th= 0.3, power threshold is P_{th min}= Noise power -6dB, P_{th max}= Noise power +2 dB.
[0076]
The spectrum of the output signal z (k, n) of the selection means 10 is shown in FIG.
This corresponds to the spectrum in the first, 50th, 100th, 150th, and 210th range bins corresponding to FIG. It can be seen that the target signal is saved and only clutter can be suppressed. The output signal selected by the selection means 10 was the second notch filter 2 nearer than the 150th range bin, and the third notch filter 3 beyond that. It was determined to be a bimodal spectrum clutter and a unimodal spectrum clutter, respectively. In the 210th range bin where the target exists, it was judged as a clutter of a unimodal spectrum, and as a result, the clutter could be suppressed without erasing the target signal. It is emphasized that this process does not require several antenna scans and is performed on the received signal within one coherent processing interval.
[0077]
【The invention's effect】
As described above, according to the present invention, since the feature quantity used in the determination unit can be calculated from the received signal within one coherent processing interval, the distribution related to the actual clutter distance without requiring several scans. It is possible to suppress clutter by adapting to.
[0078]
In addition, since the signal to be selectively output by the selection means is determined for each block composed of a plurality of continuous range bins, clutter suppression can be performed in a short time.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a clutter suppression device according to Embodiment 1 of the present invention;
FIG. 2 is a diagram schematically illustrating a clutter center frequency estimation method according to Document 2. FIG.
FIG. 3 is a spectrum of a signal including a target signal and both a fixed clutter and a moving clutter generated artificially by a computer.
4 is a diagram showing the relationship between the average value of the absolute values of the poles of the range bin and the AR model with respect to the spectrum signal of FIG. 3;
FIG. 5 is a flowchart showing a second determination procedure in the determination unit 20;
FIG. 6 is a configuration diagram of a clutter suppression device according to Embodiment 2 of the present invention;
FIG. 7 is a configuration diagram of a clutter suppression device according to Embodiment 3 of the present invention and capable of dealing with three or more clutters.
FIG. 8 is another configuration diagram of a clutter suppression device according to Embodiment 3 of the present invention and capable of dealing with three or more clutters.
FIG. 9 is an explanatory diagram of a spectrum for a specific range bin of a signal obtained as a result of clutter suppression processing according to Embodiment 2 of the present invention;
FIG. 10 is a conceptual configuration diagram of a conventional clutter suppression device.
11 is a configuration diagram of a clutter suppression device disclosed in Document 1. FIG.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 1st notch filter, 2nd 2nd notch filter, 3rd notch filter, 4th I notch filter, 5 Clutter center frequency estimation means, 10 Selection means, 20 Determination means, 30-1 Notch filter # 1 30-2-1 Notch filter # 2-1, 30-2-2 Notch filter # 2-2, 30-I-1 Notch filter # I-1, 30-I-2 Notch filter # I-2, 30 -I-I Notch filter # I-I.

Claims (4)

In a clutter suppression device that suppresses clutter, which is an unnecessary reflected echo received by a radar,
A clutter center frequency estimating means for estimating a center frequency of clutter included in a received signal converted into a digital in-phase / quadrature signal based on a maximum entropy method on the assumption that there are a maximum of a plurality of I clutters;
A plurality of I notch filters connected in cascade with the received signal as an input;
Selecting means for receiving the received signal and each of the plurality of I notch filter output signals as input and selectively outputting one of them;
Determination means for determining, based on the poles of an AR (auto-regressive) model obtained by the clutter center frequency estimation means, a signal to be selected and output by the selection means for each block composed of a plurality of continuous range bins;
The clutter suppression device according to claim 1, wherein the plurality of I notch filters use a plurality of I clutter center frequencies estimated by the clutter center frequency estimating means as zero points of frequency characteristics. 2. The clutter suppression device according to claim 1, wherein the determination means is based on the poles of the AR model obtained by the clutter center frequency estimation means and the respective powers of the received signal and the plurality of notch filter output signals. A clutter suppression apparatus, wherein a signal to be selectively output by a selection means is determined for each block composed of a plurality of continuous range bins. In a clutter suppression device that suppresses clutter, which is an unnecessary reflected echo received by a radar,
A clutter center frequency estimating means for estimating a center frequency of clutter contained in a received signal converted into a digital in-phase / quadrature signal based on a maximum entropy method on the assumption that one to a plurality of I clutters exist;
A total of I (i = 1, 2,..., I) cascaded notch filters connected in parallel with the received signal as an input to i = 1, 2,. (I + 1) / 2 notch filters;
The received signal and a plurality of I output signals at the last stage of the i (i = 1, 2,..., I) cascade-connected notch filters are input, and one of the signals is selectively output. A selection means;
Determination means for determining, for each block composed of a plurality of continuous range bins, a signal to be selected and output by the selection means based on the pole of the AR model obtained by the clutter center frequency estimation means;
The frequency that is the zero point of each of the i (i = 1, 2,..., I) cascade-connected notch filters is estimated by the clutter center frequency estimating means when it is assumed that there are i clutters. Clutter suppression device characterized in that it is a value. 4. The clutter suppression device according to claim 3, wherein the determination means includes the poles of the AR model obtained by the clutter center frequency estimation means and the received signals and i (i = 1, 2,..., I). Based on the power of each of a plurality of I output signals at the last stage of the cascade-connected notch filter, a signal to be selected and output by the selection unit is determined for each block composed of a plurality of continuous range bins. Clutter suppression device.

Images