JP3627138B2 - Noise whitening method - Google Patents

Description

【００１３】
また、ビーム出力中で信号が雑音の暴れの中からどれだけ浮き上がってくるかを表す指標である信号検出指標ｄは、以下の式で定義される。
【００１４】
ｄ＝Ｐ_Ｓ ^２／σ_Ｎ ^２ …（３）
また、積分処理出力のＳ／Ｎは、Ｓ／Ｎ_ＯＵＴ ＝５ｌｏｇｄで与えられる。
ここで、簡単な場合として、平面配列のアレイを考え、簡単のため、信号成分と背景雑音との受波器入力が相似のスペクトル特性を持ち、白色化フィルタ２−１〜２−Ｌにより白色化されているとする。
【００１５】
白色化フィルタ２−１〜２−Ｌのフィルタ出力は、Ａ／Ｄ変換手段３−１〜３−ＬによりＡ／Ｄ変換された後、ビーム出力生成手段４によりビーム形成処理が行われる。
このとき、ビーム形成による雑音の抑圧度は周波数により異なるが、雑音が３次元等方性であり、アレイの寸法が波長に比べて大きい場合には、周波数の２乗にほぼ比例する抑圧度が得られる。
【００１６】
ここで、受波器出力の信号と雑音のスペクトルをそれぞれ、Ｓ_０ 、Ｎ_０ とおくと、所望信号の到来方向に向けられたビーム出力の信号成分Ｓ（ｆ） 及び雑音成分のスペクトル特性Ｎ（ｆ） はそれぞれ、以下の式で表される。
Ｓ（ｆ） ＝Ｓ_０
Ｎ（ｆ） ＝ａ（ｆ_１／ｆ）^２ Ｎ_０ …（４）
但し、ａは定数である。
【００１７】
この信号と雑音について、積分出力の検出指標ｄを求めてみる。
以下に示すように、周波数窓Ｇ（ｆ） が平坦な場合Ｇ_０（ｆ）と、雑音成分を白色化するように周波数の２乗に比例する特性を持たせた場合Ｇ_１（ｆ）とを考える。
Ｇ_０（ｆ）＝１
Ｇ_１（ｆ）＝ｂ（ｆ／ｆ_１）^２ …（５）
【００１８】
そして、上記（４）式、（５）式を、（１）式、（２）式に代入し、（３）式を用いると、それぞれの場合について、信号検出指標ｄ_０ 、ｄ_１ が以下に示したように求められる。
【００１９】
【数２】

【００２０】
また、（６）式の両者の比を取ると以下の式が得られる。
【００２１】
【数３】

【００２２】
ここで、例えば、周波数帯域が３オクターブ（ｆ_２ ＝８ｆ_１ ）の場合を考えると、周波数窓Ｇ_０（ｆ）とＧ_１（ｆ）との違いによる出力の検出指標の差は、以下に示すようになり、周波数窓の選び方によって信号検出性能に大きな差が生じることが分かる。
５ＬＯＧ（ｄ_１／ｄ_０）＝ ７．２（ｄＢ） …（８）
【００２３】
このように、従来の雑音白色化方法では、ビーム形成、積分処理によるＳ／Ｎ改善効果が積分出力の雑音スペクトル特性に左右され、雑音のスペクトル特性が白色化フィルタや周波数窓の設計における過程からはずれている場合に所望の信号検出性能が得られないという問題点があった。
【００２４】
【課題を解決するための手段】
本発明に係る雑音白色化方法は、
アレイ型センサで受信された受波器信号の雑音成分のスペクトル特性をそれぞれ白色化する雑音白色化工程と、雑音白色化工程の出力信号をそれぞれアナログ／デジタル変換するアナログ／デジタル変換工程と、アナログ／デジタル変換工程の出力信号に基づいて、複数の方位のビーム出力を形成するビーム形成工程と、ビーム形成工程により生成されたビーム出力をあらかじめ設定された分析幅で周波数分析する周波数分析工程と、周波数分析工程により周波数分析された周波数分析結果及びあらかじめ設定された正規化を行う周波数別の空間幅の情報に基づいて、雑音成分のスペクトル特性を推定する雑音スペクトル推定工程と、周波数分析工程により周波数分析された周波数分析結果及び雑音スペクトル推定工程により推定された雑音成分のスペクトルの推定値に基づいて、各ビームの周波数成分を雑音成分のスペクトルの推定値で正規化するスペクトル正規化工程と、スペクトル正規化工程により正規化されたスペクトルを帯域について加算する積分処理工程と、積分処理工程による加算結果を処理部に出力する出力工程とを備えるものである。
【００２５】
【発明の実施の形態】
実施の形態１．
この実施の形態は、ビーム出力における雑音スペクトル特性を全ビームのスペクトルの測定値から推定し、推定した雑音スペクトル特性を用いて各ビーム出力スペクトルを正規化し、積分出力の信号検出指標の最大化を図るようにしたものである。
【００２６】
図１は本発明の一実施の形態に係る雑音白色化方法の処理の流れを説明するための説明図である。
図において、１−１〜１−Ｌはアレイ型センサの受波器信号が入力される受波信号入力端子、２−１〜２−Ｌは受波信号入力端子１−１〜１−Ｌから入力された受波器信号の雑音スペクトル特性をそれぞれ白色化するための白色化フィルタ、３−１〜３−Ｌは白色化フィルタ２−１〜２−Ｌの出力信号をそれぞれＡ／Ｄ変換するＡ／Ｄ変換手段、４はＡ／Ｄ変換手段３−１〜３−Ｌの出力信号が入力されビーム出力を形成するビーム形成手段、７はビーム出力の周波数分析を行う周波数分析手段、８は各周波数成分の方位分布から、雑音成分のスペクトル特性を推定する雑音スペクトル推定手段、９は各ビームの周波数成分を雑音スペクトルの推定値で正規化するスペクトル正規化手段、１０は正規化されたスペクトルを帯域について加算し積分処理を行う積分処理手段、１４は積分処理手段１０の出力信号を表示や信号検出、方位推定処理へ出力する出力端子である。
【００２７】
次に、この実施の形態の動作について説明する。
まず、受波信号入力端子１−１〜１−Ｌから入力されたＬ個の受波器の信号は、雑音成分を白色化するための白色化フィルタ２−１〜２−Ｌを通り、Ａ／Ｄ変換手段３−１〜３−Ｌに送られてＡ／Ｄ変換される。
【００２８】
そして、Ａ／Ｄ変換手段３−１〜３−ＬによりＡ／Ｄ変換された各受波器の出力は、ビーム生成手段４に送られ、複数の方位（方位数をＭとする）のビーム出力を生成する処理が行われる。
そして、ビーム生成手段４により生成されたＭ個の方位ビーム出力は、周波数分析手段７に送られ、あらかじめ設定された分析幅で周波数分析され、各ビームの周波数スペクトルＸ（ｆ_ｋ，θ_ｍ） 、ｋ＝１，．．．，Ｋ、ｍ＝１，．．．，Ｍが算出される。
ここで、ｆ_ｋ（ｋ＝１，．．．，Ｋ）は周波数分析におけるサンプル周波数である。
この周波数分析の方法としては、ＦＦＴや並列に配置されたフィルタ群などがある。
【００２９】
そして、周波数分析手段７による周波数分析により得られた各周波数のスペクトルは雑音スペクトル推定手段８及びスペクトル正規化手段９に送られる。
そして、雑音スペクトル推定手段８では、周波数分析手段７から送られた周波数分析結果をもとに、各周波数について、雑音スペクトルＮ（ｆ_ｋ）、ｋ＝１，．．．，Ｋを推定する。
この雑音スペクトル推定の方法としては、全ビーム出力の平均値を用いる方法、中央値（メジアン）を用いる方法などがある。
【００３０】
そして、推定された雑音スペクトルＮ（ｆ_ｋ）、ｋ＝１，．．．，Ｋはスペクトル正規化手段９に送られ、スペクトル正規化手段９では、雑音スペクトル推定手段８から雑音スペクトル推定値Ｎ（ｆ_ｋ）、ｋ＝１，．．．，Ｋが送られると、以下に示すように、周波数分析手段７から送られた各ビーム出力の周波数スペクトルＸ（ｆ_ｋ，θ_ｍ） との比を正規化されたビーム出力スペクトルＹ（ｆ_ｋ，θ_ｍ）として計算し、積分処理手段１０に送出する。

【００３１】
そして、積分処理手段１０では、スペクトル正規化手段９から送られた正規化されたビーム出力スペクトルをビーム毎に全帯域、ｆ_１〜ｆ_Ｋについて加算し、出力端子１４から表示処理や信号検出、到来方位推定などの処理に送出する。
ここで、帯域加算した各ビームの出力は、所定の時間だけ積分するようにしてもよい。
【００３２】
この実施の形態では、雑音スペクトル特性を全ビームのスペクトル特性を用いて推定し、それに基づいて、各ビーム出力のスペクトル特性を正規化しているため、雑音スペクトル特性の仮定からのズレや時間的な変動がある場合にも、レベルの高い雑音周波数成分に影響されることなく、安定に背景雑音の白色化が行われ、信号検出や方位推定に最適な帯域積分出力を供給することが可能となる。
【００３３】
実施の形態２．
この実施の形態は、雑音が全方位一様ではなく、緩やかな方位特性を持つ場合に、アレイのビーム幅に応じて空間的に局所的な正規化を行うようにしたものであり、離れた方位に高い雑音や強い別の信号源がある場合に、それに影響されずに信号を検出できるようにしたものである。
【００３４】
図２は本発明の他の実施の形態に係る雑音白色化方法の処理の流れを説明するための説明図である。
図において、１−１〜１−Ｌはアレイ型センサの受波器信号が入力される受波信号入力端子、２−１〜２−Ｌは受波信号入力端子１−１〜１−Ｌから入力された受波器信号の雑音スペクトル特性をそれぞれ白色化するための白色化フィルタ、３−１〜３−Ｌは白色化フィルタ２−１〜２−Ｌの出力信号をそれぞれＡ／Ｄ変換するＡ／Ｄ変換手段、４はＡ／Ｄ変換手段３−１〜３−Ｌの出力信号が入力されビーム出力を形成するビーム形成手段、７はビーム出力の周波数分析を行う周波数分析手段、１１は正規化を行う空間幅を記憶した正規化幅テーブル、１２は空間的に局所的な雑音スペクトルを推定するローカルな雑音スペクトル推定手段、１３は空間的に異なった雑音レベル値を用いて正規化を行うスペクトル正規化手段、１０は正規化されたスペクトルを帯域について加算し積分処理を行う積分処理手段、１４は積分処理手段１０の出力信号を表示や信号検出、方位推定処理へ出力する出力端子である。
【００３５】
次に、この実施の形態の動作について説明する。
まず、受波信号入力端子１−１〜１−Ｌから入力されたＬ個の受波器の信号は、雑音成分を白色化するための白色化フィルタ２−１〜２−Ｌを通り、Ａ／Ｄ変換手段３−１〜３−Ｌに送られてＡ／Ｄ変換される。
【００３６】
そして、Ａ／Ｄ変換手段３−１〜３−ＬによりＡ／Ｄ変換された各受波器の出力は、ビーム生成手段４に送られ、複数の方位（方位数をＭとする）のビーム出力を生成する処理が行われる。
そして、ビーム生成手段４により生成されたＭ個の方位ビーム出力は、周波数分析手段７に送られ、あらかじめ設定された分析幅で周波数分析され、各ビームの周波数スペクトルＸ（ｆ_ｋ，θ_ｍ） 、ｋ＝１，．．．，Ｋ、ｍ＝１，．．．，Ｍが算出される。
ここで、ｆ_ｋ（ｋ＝１，．．．，Ｋ）は周波数分析におけるサンプル周波数である。
この周波数分析の方法としては、ＦＦＴや並列に配置されたフィルタ群などがある。
【００３７】
そして、周波数分析手段７による周波数分析により得られた各ビームの周波数スペクトルＸ（ｆ_ｋ，θ_ｍ） 、ｋ＝１，．．．，Ｋ、ｍ＝１，．．．，Ｍはローカルな雑音スペクトル推定手段１２及びスペクトル正規化手段１３に送られる。
そして、雑音スペクトル推定手段１２では周波数分析手段７から送られた周波数分析結果をもとに、各周波数について、ビーム別に雑音スペクトルＮ（ｆ_ｋ，θ_ｍ） 、ｋ＝１，．．．，Ｋ、ｍ＝１，．．．，Ｍを推定する。
このとき、全空間のビーム出力を用いるのではなく、正規化幅テーブル１１に示された周波数別の空間範囲のビーム出力データを用いる。
【００３８】
ここで、円筒型アレイのように、ビーム形成に用いる受波器配列のビーム主軸方位に対する形状が方位によらず等しい場合には、正規化に用いるビーム出力データの幅は方位に依存しない値とすることが出来るが、平面型アレイの場合や複雑な配列構造のアレイの場合には、方位によってビーム幅が異なる。
このような場合、正規化に用いるビーム出力データの範囲は方位の関数として設定することもできる。一般に、ビーム幅の数倍以上の正規化幅が取られる。
また、雑音スペクトル推定方法としては、実施の形態１と同様に、正規化幅の範囲のビーム出力の平均値を用いる方法、中央値（メジアン）を用いる方法などがある。
【００３９】
そして、雑音スペクトル推定手段１２により推定された雑音スペクトルＮ（ｆ_ｋ，θ_ｍ） 、ｋ＝１，．．．，Ｋ、ｍ＝１，．．．，Ｍはスペクトル正規化手段１３に送られ、スペクトル正規化手段１３では、雑音スペクトル推定手段１２から雑音スペクトル推定値Ｎ（ｆ_ｋ，θ_ｍ） 、ｋ＝１，．．．，Ｋ、ｍ＝１，．．．，Ｍを送られると、以下に示すように、周波数分析手段７から送られた各ビーム出力の周波数スペクトルＸ（ｆ_ｋ，θ_ｍ） との比を正規化されたビーム出力スペクトルＹ（ｆ_ｋ，θ_ｍ）として計算し、積分処理手段１０に送出する。

【００４０】
この実施の形態では、雑音スペクトル特性をビーム毎にその近傍のビームのスペクトル特性を用いて推定し、それに基づいて各ビーム出力のスペクトル特性を正規化しているため、全空間における雑音スペクトル特性の仮定からのズレや時間的な変動だけでなく、空間的に限られた方位に高いレベルの雑音がある場合にも、それに影響されることなく、ローカルに背景雑音の白色化が行われ、信号検出や方位推定に最適な帯域積分出力を供給することが可能となる。
特に、離れた方位にある高いレベルの広帯域雑音や広帯域目標の影響を排除して、低いレベルの目標を強調して捉えるのに有効である。
【００４１】
なお、実施の形態１、実施の形態２では、海洋中における音響信号の検出に適用する場合の例として説明しているが、本発明は海洋音響に限らず、雑音の中から広帯域の信号をアレイ型センサで検出し、方位推定を行う全ての問題に適用することができる。
【００４２】
【発明の効果】
以上のように本発明によれば、アレイ型センサで受信された受波器信号の雑音成分のスペクトル特性をそれぞれ白色化し、その出力信号をそれぞれアナログ／デジタル変換し、その出力信号に基づいて、複数の方位のビーム出力を形成し、その生成されたビーム出力をあらかじめ設定された分析幅で周波数分析し、周波数分析された周波数分析結果及びあらかじめ設定された正規化を行う周波数別の空間幅の情報に基づいて、雑音成分のスペクトル特性を推定し、周波数分析された周波数分析結果及び推定された雑音成分のスペクトルの推定値に基づいて、各ビームの周波数成分を雑音成分のスペクトルの推定値で正規化し、正規化されたスペクトルを帯域について加算し、その加算結果を処理部に出力するようにしたので、雑音スペクトル特性の仮定からのズレや時間的な変動がある場合にも、レベルの高い雑音周波数成分に影響されることなく、安定に背景雑音の白色化が行われ、信号検出や方位推定に最適な帯域積分出力を供給することができるという効果を有する。
【図面の簡単な説明】
【図１】本発明の一実施の形態に係る雑音白色化方法の処理の流れを説明するための説明図である。
【図２】本発明の他の実施の形態に係る雑音白色化方法の処理の流れを説明するための説明図である。
【図３】従来の雑音白色化方法の処理の流れを説明するための説明図である。
【符号の説明】
１−１〜１−Ｌ 受波信号入力端子
２−１〜２−Ｌ 白色化フィルタ
３−１〜３−Ｌ Ａ／Ｄ変換手段
４ ビーム形成手段
７ 周波数分析手段
８ 雑音スペクトル推定手段
９ スペクトル正規化手段
１０ 積分処理手段
１１ 正規化幅テーブル
１２ ローカルな雑音スペクトル推定手段
１３ スペクトル正規化手段
１４ 出力端子[0001]
BACKGROUND OF THE INVENTION
The present invention, for example, receives a wideband signal arriving from a distance under a noise background using an array type sensor, and performs noise whitening to whiten a spectrum characteristic of noise when detecting a signal and estimating an arrival direction. It relates to the conversion method.
[0002]
[Prior art]
Conventionally, when estimating a signal in the ocean, a beam is formed by an array type sensor composed of a plurality of sensor elements to improve a signal-to-noise ratio (hereinafter referred to as S / N) and estimate an arrival direction. There are many cases.
FIG. 3 is an explanatory diagram for explaining the flow of processing of the conventional noise whitening method, and shows the flow of processing of broadband signal detection and azimuth detection using an array type sensor.
[0003]
In the figure, 1-1 to 1-L are received signal input terminals to which the receiver signals of the array type sensor are input, and 2-1 to 2-L are received from the received signal input terminals 1-1 to 1-L. Whitening filters for whitening the noise spectrum characteristics of the received receiver signal, and 3-1 to 3-L A / D convert the output signals of the whitening filters 2-1 to 2-L, respectively. A / D conversion means, 4 is a beam forming means for forming a beam output when the output signals of the A / D conversion means 3-1 to 3-L are inputted, 5 is an integration processing means for performing band addition and time integration, and 6 is This is an output terminal for outputting the output signal of the integration processing means 5 to display, signal detection, and azimuth estimation processing.
[0004]
Next, a conventional noise whitening method will be described.
First, the signals of the L receivers input from the reception signal input terminals 1-1 to 1-L pass through the whitening filters 2-1 to 2-L for whitening the noise component, and / D conversion means 3-1 to 3-L for A / D conversion.
[0005]
Here, in general, the spectral characteristics of underwater noise are -5 dB / oct. To -6 dB / oct. In many cases, for example, the underwater noise spectrum is -6 dB / oct. Assuming that, the filter for whitening this noise can be realized by a first-order high-pass filter.
Thereby, the required dynamic range in the A / D conversion in the A / D conversion means 3-1 to 3-L can be kept low.
[0006]
Then, the output of each receiver that has been A / D converted by the A / D conversion means 3-1 to 3 -L is sent to the beam generation means 4, and is a beam having a plurality of directions (the number of directions is M). Processing to generate output is performed.
The beam output generated by the beam generating means 4 is sent to the integration processing means 5, subjected to square integration processing by applying an appropriate frequency window, and then subjected to display processing, signal detection, and arrival direction from the output terminal 6. It is sent to processing such as estimation.
[0007]
Here, as a method of beam forming and integration processing, a beam output time series is generated by delay addition in the time domain, and then square integration is performed over a frequency window. There is also a method of performing frequency analysis, performing phase correction corresponding to a time delay for each frequency component, obtaining an addition output of all receivers, and band-adding the result by weighting corresponding to the frequency window.
[0008]
[Problems to be solved by the invention]
In the conventional noise whitening method as described above, noise is whitened before A / D conversion, and then a beam is formed to calculate the beam output power.
The addition of each receiver output in beam forming is usually weighted so that the sensitivity to the incoming signal from the beam principal axis direction is constant. In this case, the noise spectrum characteristic of the beam output is from white. In many cases, it is off.
This is because even if the spectral characteristic of the receiver output is whitened by the whitening filter, the spectral characteristic is changed from white by beam forming.
[0009]
Normally, the frequency window in beam forming is set so that the frequency characteristic of the main axis sensitivity is equal to that of a single channel. In this case, the suppression amount of the noise component is larger at a high frequency with a short wavelength.
Therefore, the noise component in the low frequency band with a small suppression amount has a large influence on the integrated output, and the S / N improvement effect due to beam formation as a whole is not sufficiently exhibited.
[0010]
Even when the spectral characteristics of the noise are different from the characteristics assumed in the design of the whitening filter or the frequency window, the noise band that remains strongly due to the assumed deviation greatly affects the noise level of the integrated output, and similarly S / N improvement effect cannot be obtained as expected.
[0011]
Next, the details of the phenomenon that the S / N improvement effect cannot be obtained as expected will be described below using a planar array as an example.
First, the power spectra of the signal component and noise component of the beam output are S (f) and N (f), the power response characteristic of the frequency window in the band integration is G (f), the integration time is T, and the processing band is (f1). , placing a f2), variance sigma _N ² of the power of the power P _S and a noise component of the signal component of the beam output is respectively given by the following equations.
[0012]
[Expression 1]

[0013]
In addition, a signal detection index d, which is an index indicating how much the signal rises during noise output during beam output, is defined by the following equation.
[0014]
d = P _S ² / σ _N ² (3)
The S / N of the integration processing output is given by S / N _OUT = 5 logd.
Here, as a simple case, a planar array is considered. For simplicity, the receiver inputs of the signal component and the background noise have similar spectral characteristics, and are whitened by the whitening filters 2-1 to 2-L. Suppose that
[0015]
The filter outputs of the whitening filters 2-1 to 2-L are A / D converted by the A / D conversion units 3-1 to 3-L, and then subjected to beam forming processing by the beam output generation unit 4.
At this time, although the degree of suppression of noise due to beam formation varies depending on the frequency, when the noise is three-dimensional isotropic and the size of the array is larger than the wavelength, the degree of suppression is approximately proportional to the square of the frequency. can get.
[0016]
Here, each of the spectra of the reception output of the signal and noise, S _0, by placing the N _0, the spectral characteristics N of the signal component S (f) and the noise component of the beam output directed to the arrival direction of the desired signal (F) is respectively represented by the following formula.
S (f) = S ₀
N (f) = a (f ₁ / f) ² N ₀ (4)
However, a is a constant.
[0017]
For this signal and noise, the detection index d of the integrated output is obtained.
As shown below, G ₀ (f) when the frequency window G (f) is flat, and G ₁ (f) when a characteristic proportional to the square of the frequency is given so that the noise component is whitened. think of.
G ₀ (f) = 1
G ₁ (f) = b (f / f ₁ ) ² (5)
[0018]
Then, when the above expressions (4) and (5) are substituted into the expressions (1) and (2) and the expression (3) is used, the signal detection indices d ₀ and d ₁ are as follows in each case: It is required as shown in
[0019]
[Expression 2]

[0020]
Moreover, the following formula | equation will be obtained if both ratio of (6) Formula is taken.
[0021]
[Equation 3]

[0022]
Here, for example, considering the case where the frequency band is 3 octaves (f ₂ = 8f ₁ ), the difference in the output detection index due to the difference between the frequency windows G ₀ (f) and G ₁ (f) is as follows: It can be seen that there is a large difference in signal detection performance depending on how the frequency window is selected.
_{5LOG (d 1 / d 0)} = 7.2 (dB) ... (8)
[0023]
As described above, in the conventional noise whitening method, the S / N improvement effect by beam forming and integration processing depends on the noise spectrum characteristics of the integrated output, and the noise spectrum characteristics are obtained from the process of designing the whitening filter and frequency window. There is a problem in that a desired signal detection performance cannot be obtained when they are off.
[0024]
[Means for Solving the Problems]
The noise whitening method according to the present invention includes:
A noise whitening process for whitening the spectral characteristics of the noise component of the receiver signal received by the array sensor, an analog / digital conversion process for converting the output signal of the noise whitening process from analog to digital, and analog / A beam forming step for forming a beam output in a plurality of directions based on an output signal of the digital conversion step, and a frequency analysis step for analyzing the frequency of the beam output generated by the beam forming step with a preset analysis width, A noise spectrum estimation step for estimating a spectral characteristic of a noise component based on a frequency analysis result obtained by frequency analysis by the frequency analysis step and information on a spatial width for each frequency for which normalization is performed in advance, and a frequency by the frequency analysis step. Analyzed frequency analysis result and noise estimated by noise spectrum estimation process Spectral normalization process that normalizes the frequency component of each beam with the estimated value of the noise component spectrum based on the estimated value of the minute spectrum, and integration processing that adds the spectrum normalized by the spectral normalization process for the band A process and an output process for outputting the addition result of the integration process to the processing unit.
[0025]
DETAILED DESCRIPTION OF THE INVENTION
Embodiment 1 FIG.
In this embodiment, the noise spectrum characteristic at the beam output is estimated from the measured values of the spectrum of all the beams, each beam output spectrum is normalized using the estimated noise spectrum characteristic, and the signal detection index of the integrated output is maximized. It is intended to be illustrated.
[0026]
FIG. 1 is an explanatory diagram for explaining a processing flow of a noise whitening method according to an embodiment of the present invention.
In the figure, 1-1 to 1-L are received signal input terminals to which the receiver signals of the array type sensor are input, and 2-1 to 2-L are received from the received signal input terminals 1-1 to 1-L. Whitening filters for whitening the noise spectrum characteristics of the received receiver signal, and 3-1 to 3-L A / D convert the output signals of the whitening filters 2-1 to 2-L, respectively. A / D conversion means, 4 is a beam forming means for forming a beam output when the output signals of the A / D conversion means 3-1 to 3-L are inputted, 7 is a frequency analysis means for performing frequency analysis of the beam output, and 8 is Noise spectrum estimating means for estimating the spectral characteristics of the noise component from the azimuth distribution of each frequency component, 9 is a spectrum normalizing means for normalizing the frequency component of each beam with the estimated value of the noise spectrum, and 10 is a normalized spectrum. For the bandwidth Integration processing means for performing partial processing, 14 denotes an output terminal for outputting an output signal of the integrating unit 10 displays and signal detection, the direction estimation processing.
[0027]
Next, the operation of this embodiment will be described.
First, the signals of the L receivers input from the reception signal input terminals 1-1 to 1-L pass through the whitening filters 2-1 to 2-L for whitening the noise component, and / D conversion means 3-1 to 3-L for A / D conversion.
[0028]
Then, the output of each receiver that has been A / D converted by the A / D conversion means 3-1 to 3 -L is sent to the beam generation means 4, and is a beam having a plurality of directions (the number of directions is M). Processing to generate output is performed.
The M azimuth beam outputs generated by the beam generating unit 4 are sent to the frequency analyzing unit 7 and subjected to frequency analysis with a preset analysis width, and the frequency spectrum X (f _k , θ _m ) of each beam. , K = 1,. . . , K, m = 1,. . . , M are calculated.
Here, f _k (k = 1,..., K) is a sample frequency in the frequency analysis.
Examples of the frequency analysis method include FFT and a group of filters arranged in parallel.
[0029]
Then, the spectrum of each frequency obtained by the frequency analysis by the frequency analysis means 7 is sent to the noise spectrum estimation means 8 and the spectrum normalization means 9.
Then, the noise spectrum estimation unit 8 uses the noise spectrum N (f _k ), k = 1,. . . , K.
As a method for estimating the noise spectrum, there are a method using an average value of all beam outputs, a method using a median (median), and the like.
[0030]
Then, the estimated noise spectrum N (f _k ), k = 1,. . . , K are sent to the spectrum normalization means 9, where the noise spectrum estimation value N (f _k ), k = 1,. . . , K are sent, the beam output spectrum Y (f _k) obtained by normalizing the ratio with the frequency spectrum X (f _k , θ _m ) of each beam output sent from the frequency analysis means 7 as shown below. , Θ _m ) and sent to the integration processing means 10.

[0031]
Then, the integration processing means 10 adds the normalized beam output spectrum sent from the spectrum normalization means 9 for each beam for all bands, f _{1 to} f _K , and performs display processing and signal detection from the output terminal 14. It is sent to processing such as arrival direction estimation.
Here, the output of each beam subjected to band addition may be integrated for a predetermined time.
[0032]
In this embodiment, the noise spectral characteristics are estimated using the spectral characteristics of all the beams, and the spectral characteristics of each beam output are normalized based on the estimated spectral characteristics. Even if there is a fluctuation, the background noise is stably whitened without being affected by high-level noise frequency components, and it is possible to supply an optimal band integration output for signal detection and orientation estimation .
[0033]
Embodiment 2. FIG.
In this embodiment, when the noise is not uniform in all directions but has a gentle direction characteristic, spatially local normalization is performed according to the beam width of the array. When there is a high noise or another strong signal source in the azimuth, the signal can be detected without being affected by it.
[0034]
FIG. 2 is an explanatory diagram for explaining the flow of processing of the noise whitening method according to another embodiment of the present invention.
In the figure, 1-1 to 1-L are received signal input terminals to which the receiver signals of the array type sensor are input, and 2-1 to 2-L are received from the received signal input terminals 1-1 to 1-L. Whitening filters for whitening the noise spectrum characteristics of the received receiver signal, and 3-1 to 3-L A / D convert the output signals of the whitening filters 2-1 to 2-L, respectively. A / D conversion means, 4 is a beam forming means for receiving the output signals of the A / D conversion means 3-1 to 3-L to form a beam output, 7 is a frequency analysis means for analyzing the frequency of the beam output, and 11 is Normalization width table storing spatial width for normalization, 12 is a local noise spectrum estimation means for estimating spatially local noise spectrum, and 13 is normalized by using spatially different noise level values. Spectral normalization means to perform, 10 is normalization Integrating unit for performing integration processing spectrum by adding the band, 14 denotes an output terminal for outputting an output signal of the integrating unit 10 displays and signal detection, the direction estimation processing.
[0035]
Next, the operation of this embodiment will be described.
First, the signals of the L receivers input from the reception signal input terminals 1-1 to 1-L pass through the whitening filters 2-1 to 2-L for whitening the noise component, and / D conversion means 3-1 to 3-L for A / D conversion.
[0036]
Then, the output of each receiver that has been A / D converted by the A / D conversion means 3-1 to 3 -L is sent to the beam generation means 4, and is a beam having a plurality of directions (the number of directions is M). Processing to generate output is performed.
The M azimuth beam outputs generated by the beam generating unit 4 are sent to the frequency analyzing unit 7 and subjected to frequency analysis with a preset analysis width, and the frequency spectrum X (f _k , θ _m ) of each beam. , K = 1,. . . , K, m = 1,. . . , M are calculated.
Here, f _k (k = 1,..., K) is a sample frequency in the frequency analysis.
Examples of the frequency analysis method include FFT and a group of filters arranged in parallel.
[0037]
Then, the frequency spectrums X (f _k , θ _m ), k = 1,. . . , K, m = 1,. . . , M are sent to local noise spectrum estimation means 12 and spectrum normalization means 13.
Then, the noise spectrum estimation means 12 uses the noise spectrum N (f _k , θ _m ), k = 1,... For each frequency for each frequency based on the frequency analysis result sent from the frequency analysis means 7. . . , K, m = 1,. . . , M is estimated.
At this time, the beam output data of the spatial range for each frequency shown in the normalized width table 11 is used instead of using the beam output of the entire space.
[0038]
Here, when the shape of the receiver array used for beam formation with respect to the beam principal axis direction is the same regardless of the direction, such as a cylindrical array, the width of the beam output data used for normalization is a value independent of the direction. However, in the case of a planar array or an array with a complicated arrangement structure, the beam width differs depending on the orientation.
In such a case, the range of beam output data used for normalization can also be set as a function of azimuth. In general, a normalized width of several times the beam width is taken.
As the noise spectrum estimation method, there are a method using an average value of beam outputs in a range of a normalization width, a method using a median (median), and the like, as in the first embodiment.
[0039]
Then, the noise spectrum N (f _k , θ _m ), k = 1,. . . , K, m = 1,. . . , M are sent to the spectrum normalizing means 13, where the noise spectrum estimation value N (f _k , θ _m ), k = 1,. . . , K, m = 1,. . . , M, the beam output spectrum Y (f _k) obtained by normalizing the ratio of each beam output to the frequency spectrum X (f _k , θ _m ) sent from the frequency analysis means 7 as shown below. , Θ _m ) and sent to the integration processing means 10.

[0040]
In this embodiment, the noise spectral characteristics are estimated for each beam by using the spectral characteristics of neighboring beams, and the spectral characteristics of each beam output are normalized based on the estimated spectral characteristics. In addition to deviation from time and temporal variations, even when there is a high level of noise in a spatially limited direction, the background noise is whitened locally without being affected by it, and signal detection In addition, it is possible to supply an optimum band integration output for azimuth estimation.
In particular, it is effective for exaggerating and capturing low-level targets by eliminating the effects of high-level broadband noise and broadband targets in remote directions.
[0041]
Although the first and second embodiments have been described as an example of application to detection of an acoustic signal in the ocean, the present invention is not limited to ocean acoustics, and a broadband signal from noise is used. The present invention can be applied to all problems that are detected by an array type sensor and the direction is estimated.
[0042]
【The invention's effect】
As described above, according to the present invention, the spectral characteristics of the noise components of the receiver signal received by the array type sensor are whitened, the output signals are each analog / digital converted, and based on the output signals, A beam output of a plurality of directions is formed, and the generated beam output is subjected to frequency analysis with a preset analysis width, and the frequency analysis result of the frequency analysis and the spatial width for each frequency for performing the normalization set in advance Based on the information, the spectral characteristics of the noise component are estimated. Based on the frequency analysis result of the frequency analysis and the estimated noise component spectrum, the frequency component of each beam is converted into the estimated noise component spectrum. Since the normalized spectrum is added for the band and the addition result is output to the processing unit, the noise spectrum Even if there is a deviation from the assumption of sexuality or temporal variation, the background noise is stably whitened without being affected by high-level noise frequency components, and the optimal band for signal detection and orientation estimation The integrated output can be supplied.
[Brief description of the drawings]
FIG. 1 is an explanatory diagram for explaining a processing flow of a noise whitening method according to an embodiment of the present invention;
FIG. 2 is an explanatory diagram for explaining a processing flow of a noise whitening method according to another embodiment of the present invention.
FIG. 3 is an explanatory diagram for explaining a processing flow of a conventional noise whitening method;
[Explanation of symbols]
1-1 to 1-L Received signal input terminals 2-1 to 2-L Whitening filters 3-1 to 3-L A / D conversion means 4 Beam forming means 7 Frequency analysis means 8 Noise spectrum estimation means 9 Spectral normality Normalization unit 10 Integration processing unit 11 Normalization width table 12 Local noise spectrum estimation unit 13 Spectrum normalization unit 14 Output terminal

Claims (1)

A noise whitening step of whitening the spectral characteristics of the noise components of the receiver signal received by the array type sensor;
An analog / digital conversion step for analog / digital conversion of the output signal of the noise whitening step;
A beam forming step of forming a beam output in a plurality of directions based on an output signal of the analog / digital conversion step;
A frequency analysis step of analyzing the frequency of the beam output generated by the beam forming step with a predetermined analysis width;
A noise spectrum estimation step of estimating a spectral characteristic of a noise component based on the frequency analysis result frequency-analyzed by the frequency analysis step and information on a spatial width for each frequency for which normalization is performed in advance ,
The frequency component of each beam is normalized with the estimated value of the noise component spectrum based on the frequency analysis result analyzed by the frequency analyzing step and the estimated value of the noise component spectrum estimated by the noise spectrum estimating step. A spectral normalization step,
An integration process step of adding the spectrum normalized by the spectrum normalization step with respect to the band;
A noise whitening method, comprising: an output step of outputting the addition result of the integration processing step to a processing unit.

Images