JP2004110404A - Mixed signal separation/extraction device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a mixed signal separation/extraction device capable of reducing a computing quantity even for a large number of sensors. <P>SOLUTION: First to n-th antennas 11<SB>1</SB>-11<SB>n</SB>, on which a plurality of arrival waves are incident respectively, are arranged at predetermined intervals, and a plurality of mixed signals constructed of a plurality of mixed incident signals from the first to n-th antennas 11<SB>1</SB>-11<SB>n</SB>are restricted within a predetermined narrow band by means of first to n-th band restriction wave filters 21<SB>1</SB>-21<SB>n</SB>. In the arrival wave estimation processing part 47, the number of arrival wave signals is estimated on the basis of a plurality of observation signals restricted in the narrow band, and the estimated number of observation signals are selected from a plurality of observation signals. In an independent component analysis part 50, the estimated number of arrival wave signals are selected/extracted according to an independent component analysis method on the basis of the estimated number of selected observation signals. <P>COPYRIGHT: (C)2004,JPO

Description

という関係式を得る。
【００３４】
まず、この式（１）から観測データの相関行列の固有値を求め、雑音電力σ^２より大きく実質的に有効な信号に関する固有値の数から到来波信号の推定数ｍを推定することができる。
【００３５】
次いで、第２の手法では、例えば文献（ＭＡＴＩ　ＭＡＸ，ＴＨＯＭＡＳ　ＫＡＩＬＡＴＨ，　’Ｄｅｔｅｃｔｉｏｎ　ｏｆ　Ｓｉｇｎａｌｓ　ｂｙ　Ｉｎｆｏｒｍａｔｉｏｎ　Ｔｈｅｏｒｅｔｉｃ　Ｃｒｉｔｅｒｉａ’，　ＩＥＥＥ　Ｔｒａｎｓａｃｔｉｏｎ　ｏｎ　Ａｃｏｕｓｔｉｃｓ，　Ｓｐｅｅｃｈ，　ａｎｄ　Ｓｉｇｎａｌ　Ｐｒｏｃｅｓｓｉｎｇ，　Ｖｏｌ．　ＡＳＳＰ−３３，　Ｎｏ．２，　Ａｐｒｉｌ　１９８５）で示す手法でＡＩＣ（Ａｋａｉｋｅ　Ｉｎｆｏｒｍａｔｉｏｎ　Ｃｒｉｔｅｒｉａ）やＭＤＬ（Ｍｉｎｉｍｕｍ　Ｄｅｓｃｒｉｐｔｉｏｎ　Ｌｅｎｇｔｈ）がある。
【００３６】
ＡＩＣ（ｋ）は、ｋ＝１〜ｎの範囲で、
【数２】

という計算を行い、最小となるｋの値を到来波信号の推定数ｋとして推定する。
【００３７】
また、ＭＤＬ（ｋ）は、ｋ＝１〜ｎの範囲で、
【数３】

という計算を行い、最小となるｋの値を到来波信号の推定数ｋとして推定する。
【００３８】
このように、到来波推定処理部４７では、以上の第１および第２の手法を用いて、到来波信号の推定数ｍ，ｋを推定しておき、推定数ｍ，ｋが（ｍ≦ｋ＜ｎ）を満足する場合には、Ｘ_{１ ， Λ ，ｎ}（ｔ）の中から（ｍ≦ｋ＜ｎ）となるようなｋ個のセンサ信号Ｘ_{１ ， Λ ，ｎ}（ｔ）を無作為に選択して観測信号ｘ_{１，・・・，ｋ}（ｔ）として独立成分分析部５０に送る。
【００３９】
また、複数のアンテナで受信された到来波信号の推定数ｋが推定できた場合、この推定数ｋよりも多く、全アンテナ数より少ない数分のセンサ数の信号を使用してブラインド信号分離を行うことにより演算量を低減するようにしてもよい。
【００４０】
なお、Ｘ_{１ ， Λ ，ｎ}（ｔ）の中から無作為にｋ個を選択する方法としては、例えば、１番目〜ｋ番目までのＸ_{１ ， Λ ， ｋ}（ｔ）をｋ個だけ順に選ぶ方法、１〜ｎの範囲のｋ個の重複しない乱数を発生させてこの乱数番目のＸ_{１ ， Λ ， ｋ}（ｔ）をｋ個だけ選ぶ方法等がある。
【００４１】
独立成分分析部５０は、観測信号ｘ_{１，・・・，ｋ}（ｔ）として送られてくるｋ個のサンプリング時系列を用いて、ブラインド信号分離のアルゴリズムにより、選択された帯域内の信号ｙ（ｔ）を分離・抽出する。
【００４２】
以下、分離・抽出の手順を詳細に説明する。
この第１の実施の形態におけるブラインド信号分離は、時間遅れなく混合された信号の分離に対して用いられる。先ず、その基本となるブラインド信号分離の問題についてここで定義する。信号源が下記式（４）のベクトルで与えられるとする。
【００４３】
【数４】

但し、ｓ（ｔ）は、ｎ個の入射信号であり、平均「０」であって、互いに独立であるとする。また、Ｔは転置を表す。
【００４４】
観測信号は、各アンテナ１１_１〜１１_ｎで観測され帯域制限された時系列データを意味しており、
【数５】

で表すものとする。これは、１，・・・，ｎの各アンテナ１１_１〜１１_ｎで観測された信号であると考えることができる。一般には、アンテナの数と信号源の数とは必ずしも一致しないが、ここでは一致しているものとする。
【００４５】
単純なＩＣＡの問題では、ｓ（ｔ）とｘ（ｔ）との間に、
【数６】

なる単純な線形関係を仮定する。Ａは、各アンテナ１１_１〜１１_ｎの配置と特性で決まる信号混合行列（ｎ行×ｎ列）の実数行列である。ｓ（ｔ）やＡに関する知識を持たずｘ（ｔ）を独立な信号成分に分離する。
【００４６】
即ち、あるｎ×ｎの実数行列を求めることにより、
【数７】

で求まる互いに独立なｙ（ｔ）を再構成することがＩＣＡの目的である。Ｂは理想的にはＡ^−１となればよいわけだが、そうはならず順番の入れ違い（ｐｅｒｍｕｔａｔｉｏｎ）と大きさ（ａｍｐｌｉｔｕｄｅ）の任意性は残ってしまう。
【００４７】
この問題の解法の１つとして、確率分布の独立性に基づく分離法がある。各ｓ_ｉ（ｔ）が（強）定常でガウシアン（Ｇａｕｓｓｉａｎ）でないという仮定のもとで、ｙ_ｉ（ｔ）が互いに独立になるようにＢを求める手法がさまざまに提案されているが、それらの多くは次のようにまとめることができる。ｙ（ｔ）を強定常過程として、その同時分布の密度関数を、
【数８】

とすると、独立性の定義はｐ（ｙ_ｉ）をｐ（ｙ）のｙ_ｉについての周辺分布として、
【数９】

とかける。
【００４８】
同時分布と周辺分布の積との間Ｋｕｌｌｂａｃｋ−Ｌｅｉｂｌｅｒ　発散（ｄｉｖｅｒｇｅｎｃｅ）は、
【数１０】

となる。但し、Ｈ（Ｙ；Ｂ）は同時分布ｐ（ｙ）のエントロピー、Ｈ（Ｙ_ｉ；Ｂ）は周辺分布ｐ（ｙ_ｉ）のエントロピーである。
【００４９】
これは｛Ｙ_ｉ｝（ｉ＝１，・・・，ｎ）の相互情報量である。信号源が正規分布でないという仮定からＫＬ（Ｂ）はｐ（ｙ_ｉ）が互いに独立な場合に限り「０」となる。これらはｐ（ｘ）とＢによって定まるものである。
【００５０】
ここで、ｐ（ｙ）ｄｙ＝ｐ（ｘ）ｄｘ、ｐ（ｙ）＝ｐ（ｘ）／｜Ｂ｜（｜Ｂ｜はＢの行列式）であることに注意すると、
【数１１】

となる。
【００５１】
一方、周辺分布のエントロピーは、
【数１２】

である。よって、
【数１３】

となり、
【数１４】

のようにすれば最急降下法として正しいＢを求めることができる。上記式（１４）の中で問題となるのは逆行列（Ｂ^Ｔ）^−１を計算している点である。
【００５２】
収束性に関しては、これにいかなる正定値行列を掛けても構わないことから、Ｂ^ＴＢを掛ければ（これは正則な行列の多様体上でのリーマン（Ｒｉｅｍａｎ）計量に対応している）、
【数１５】

が新たな学習則となる。定常性の仮定よりｐ（ｓ）、ｐ（ｘ）とｐ（ｙ）は時間的に独立である。この仮定のもと、アンサンブル平均を時間平均に置き換えることができる。
【００５３】
【数１６】

したがって、ηを正の定数とし、データが観測される毎に下記式（１７）に従ってパラメータを更新すればＢ_ｔが得られる。
【００５４】
【数１７】

ここで、当然問題になるのは、上記式（１３）のｐ（ｙ_ｉ）或いはψ（ｙ）をいかに定義するかである。通常、これはパラメトリックな非線形関数や統計的な展開法が用いられる。
【００５５】
大雑把な考え方を示すと、もしｐ（ｙ_ｉ）が正規分布ならはψ（ｙ）は線形関数となる。一方、正規分布より裾が”重い”場合（ｓｕｂ−Ｇａｕｓｓｉａｎ）多項式などで近似するのがよく、正規分布より裾が”軽い”場合（ｓｕｐｅｒ−Ｇａｕｓｓｉａｎ）シグモイド（Ｓｉｇｍｏｉｄ）関数などで近似するのがよいとされている。音声信号などは裾が”軽い”ので、シグモイド関数などがうまく働く。
以上のようにして独立成分分析部５０で分離・抽出された信号ｙ_{１，・・・，ｋ}（ｔ）は出力処理部６０に送られる。
【００５６】
出力処理部６０では、独立成分分析部５０からのデジタル信号として送られてくる信号ｙ_{１，・・・，ｋ}（ｔ）をアナログ信号に変換する。より詳しくは、出力処理部６０から出力される信号ｙ_１（ｔ）は、第１Ｄ／Ａ変換器６１でデジタル信号に変換され、分離信号Ｏ_１として外部に送出される。同様に、信号ｙ_{２，・・・，ｋ}（ｔ）は、第２〜第ｋＤ／Ａ変換器６１_２〜６１_ｋでデジタル信号にそれぞれ変換され、分離信号Ｏ_２〜Ｏ_ｋとして外部にそれぞれ送出される。
【００５７】
この結果、独立成分分析部５０ではｋ個の観測信号により独立成分分析の演算を行うので、従来のようにｎ個で演算するよりも演算量を低減でき、演算時間の短縮に寄与することができる。
【００５８】
以上説明したように、この第１の実施の形態に係る混合信号分離・抽出装置によれば、複数の到来波が各々に入射される第１〜第ｎアンテナ１１_１〜１１_ｎを所定間隔で配置しておき、第１〜第ｎアンテナ１１_１〜１１_ｎからの複数の入射信号が混合された複数の混合信号を第１〜第ｎ帯域制限ろ波器２１_１〜２１_ｎで所定の狭帯域にそれぞれ制限し、到来波推定処理部４７では狭帯域に制限された複数の観測信号に基づいて、到来波信号の数を推定して複数の観測信号の中から当該推定数分の観測信号を選択し、独立成分分析部５０では選択された推定数分の観測信号に基づいて、独立成分分析の手法により推定数分の到来波信号を分離・抽出することで、アンテナ数が多い場合でも、独立成分分析に関する演算量を低減することができ、演算時間の低減に寄与することができる。
【００５９】
（第２の実施の形態）
本発明の第２の実施の形態に係る混合信号分離・抽出装置は、センサとしてマイクロフォンを使用し、複数のマイクロフォンに入射された複数の入射信号の混合信号から原信号を分離して出力する。なお、以下では、説明を簡単にするために、アンテナの数を「ｎ」とし、入射信号の数を「ｎ」として説明するが、マイクロフォンの数及び入射信号の数はこれらに限定されず任意である。また、第１の実施の形態と同一の構成要素には同一の符号を付し、説明を省略又は簡略化する。
【００６０】
図２は、本発明の第２の実施の形態に係る混合信号分離・抽出装置の構成を示すブロック図である。
この混合信号分離・抽出装置は、第１〜第ｎマイクロフォン７１_１〜７１_ｎ、第１〜第ｎ帯域制限ろ波モジュール８１_１〜８１_ｎ、帯域制限制御器２５、サンプリング部９０、到来波推定処理部４７、独立成分分析部５０及び出力処理部６０から構成されている。
【００６１】
第１〜第ｎマイクロフォン７１_１〜７１_ｎとしては、無指向性、及び任意の指向性を持った種々のマイクロフォンを用いることができ、全方位からの音を受信する。これら第１〜第ｎマイクロフォン７１_１〜７１_ｎを設置する間隔や高さは任意である。
【００６２】
第１マイクロフォン７１_１は、空中からの複数の入射信号（音波）Ｓ_１〜_ｎを受信し、これらが混合された混合信号を第１帯域制限ろ波モジュール８１_１に送る。同様に、第２〜第ｎマイクロフォン７１_２〜７１_ｎは、空中からの複数の入射信号Ｓ_１〜Ｓ_ｎを受信し、これらが混合された混合信号を第２〜第ｎ帯域制限ろ波モジュール８１_１〜８１_ｎにそれぞれ送る。
【００６３】
第１帯域制限ろ波モジュール８１_１は、第１マイクロフォン７１_１からの混合信号を増幅するプリアンプと混合信号に含まれる所定帯域の周波数成分のみを通過させる帯域制限ろ波器とから構成されている。この第１帯域制限ろ波モジュール８１_１は、第１マイクロフォン７１_１からの混合信号を増幅し、増幅された混合信号に含まれる所定帯域の周波数成分のみを通過させてサンプリング部９０に送る。
【００６４】
同様に、第２〜第ｎ帯域制限ろ波モジュール８１_２〜８１_ｎは、第２〜第ｎマイクロフォン７１_２〜７１_ｎからの混合信号を増幅するプリアンプと混合信号に含まれる所定帯域の周波数成分のみを通過させる帯域制限ろ波器とから構成されている。これらの第２〜第４帯域制限ろ波モジュール８１_２〜８１_ｎは、第２〜第４マイクロフォン７１_２〜７１_ｎからの混合信号をそれぞれ増幅し、増幅された混合信号に含まれる所定帯域の周波数成分のみを通過させてサンプリング部９０にそれぞれ送る。なお、各第１〜第ｎ帯域制限ろ波モジュール８１_１〜８１_ｎが通過させる周波数帯域は同じである。
【００６５】
帯域制限制御器２５は、第１の実施の形態のそれと同じである。したがって、帯域制限制御器２５からの制御信号を適宜変更することにより、第１〜第４帯域制限ろ波モジュール８１_１〜８１_ｎを通過する周波数帯域を任意に変化させることができる。
【００６６】
サンプリング部９０は、第１〜第ｎ増幅器９１_１〜９１_ｎ、第１〜第ｎＡ／Ｄ変換器４１_１〜４１_ｎ及び発振器４５から構成されている。
第１〜第ｎ増幅器９１_１〜９１_ｎは、第１〜第ｎ帯域制限ろ波モジュール８１_１〜８１_ｎからの信号をそれぞれ増幅する。第１〜第ｎ増幅器９１_１〜９１_ｎで増幅された信号は、第１〜第ｎＡ／Ｄ変換器４１_１〜４１_ｎにそれぞれ送られる。第１〜第ｎＡ／Ｄ変換器４１_１〜４１_ｎ及び発振器４５の構成は、第１の実施の形態のそれらと同じである。第１〜第ｎＡ／Ｄ変換器４１_１〜４１_ｎの各々から出力されるデジタル信号は、到来波推定処理部４７に供給される。
到来波推定処理部４７、独立成分分析部５０及び出力処理部６０の構成は、第１の実施の形態のそれらと同じである。
【００６７】
次に、上記のように構成される本発明の第２の実施の形態に係る混合信号分離・抽出装置の動作を説明する。
各マイクロフォン７１_１〜７１_ｎで観測される複数の入射信号Ｓ_{１，・・・，ｎ}の混合信号ｘ_{１，・・・，ｎ}（ｔ）は、第１〜第ｎ帯域制限ろ波モジュール８１_１〜８１_ｎにそれぞれ送られる。第１〜第ｎ帯域制限ろ波モジュール８１_１〜８１_ｎは、混合信号ｘ_{１，・・・，ｎ}（ｔ）を増幅した後、所定帯域の周波数成分のみをそれぞれ通過させ、サンプリング部９０に送る。
【００６８】
サンプリング部９０の第１増幅器９１_１は、第１帯域制限ろ波モジュール８１_１からの混合信号ｘ_１（ｔ）を入力して増幅する。この第１増幅器９１で増幅された信号は第１Ａ／Ｄ変換器４１_１に送られる。
【００６９】
第１Ａ／Ｄ変換器４１_１は、第１増幅器９１_１からのアナログ信号を、発振器４５からのサンプリングクロックを用いてサンプリングすることによりデジタル信号に変換し、観測信号Ｘ_１（ｔ）として出力する。
【００７０】
同様に、第２〜第ｎ増幅器９１_１〜９１_ｎは、第２〜第ｎ帯域制限ろ波モジュール８１_２〜８１_ｎからの混合信号ｘ_{２，・・・，ｎ}（ｔ）をそれぞれ入力して増幅し、第２〜第ｎＡ／Ｄ変換器４１_２〜４１_ｎにそれぞれ送る。
【００７１】
第２〜第ｎＡ／Ｄ変換器４１_２〜４１_ｎは、第２〜第ｎ増幅器９１_２〜９１_ｎからのアナログ信号を、発振器４５からのサンプリングクロックを用いてサンプリングすることによりデジタル信号に変換し、観測信号Ｘ_{２，・・・，ｎ}（ｔ）として出力する。この際、サンプリング間隔（サンプリングクロックの周波数）は、各マイクロフォン７１_２〜７１_ｎに到達する入射信号Ｓ_{１，・・・，ｎ}の時間差の影響が出ない値に調整される。このようにしてサンプリング部９０で生成された観測信号Ｘ_{１，・・・，ｎ}（ｔ）は、到来波推定処理部４７に送られる。
【００７２】
到来波推定処理部４７では、上述した第１および第２の手法を用いて、到来波の信号数ｍ，ｋを推定しておき、（ｍ≦ｋ＜ｎ）を満足する場合には、Ｘ_{１ ， Λ ，ｎ}（ｔ）の中から（ｍ≦ｋ＜ｎ）となるようなｋ個のセンサ信号Ｘ_{１ ， Λ ，ｎ}（ｔ）を無作為に選択して観測信号ｘ_{１，・・・，ｋ}（ｔ）として独立成分分析部５０に送る。
【００７３】
独立成分分析部５０は、観測信号ｘ_{１，・・・，ｎ}（ｔ）として送られてくるｋ個のサンプリング時系列を用いて、ブラインド信号分離のアルゴリズムにより、選択された帯域内の信号ｙ（ｔ）を分離・抽出する。分離・抽出の手順は、第１の実施の形態で説明した手順と同じである。この独立成分分析部５０で分離・抽出された信号ｙ_{１，・・・，ｎ}（ｔ）は出力処理部６０に送られる。
【００７４】
出力処理部６０では、独立成分分析部５０からのデジタル信号として送られてくる信号ｙ_{１，・・・，ｋ}（ｔ）をアナログ信号に変換する。より詳しくは、出力処理部６０から出力される信号ｙ_１（ｔ）は、第１Ｄ／Ａ変換器６１でデジタル信号に変換され、分離信号Ｏ_１として外部に送出される。同様に、信号ｙ_{２，・・・，ｋ}（ｔ）は、第２〜第ｎＤ／Ａ変換器６１_１〜６１_ｋでデジタル信号にそれぞれ変換され、分離信号Ｏ_２〜Ｏ_ｋとして外部にそれぞれ送出される。
【００７５】
以上説明したように、この第２の実施の形態に係る混合信号分離・抽出装置によれば、複数の到来波が各々に入射されるマイクロフォン７１_１〜７１_ｎを所定間隔で配置しておき、第１〜第ｎマイクロフォン７１_１〜７１_ｎからの複数の入射信号が混合された複数の混合信号を第１〜第ｎ帯域制限ろ波モジュール８１_１〜８１_ｎで所定の狭帯域にそれぞれ制限し、到来波推定処理部４７では狭帯域に制限された複数の観測信号に基づいて、到来波信号の数を推定して複数の観測信号の中から当該推定数分の観測信号を選択し、独立成分分析部５０では選択された推定数分の観測信号に基づいて、独立成分分析の手法により推定数分の到来波信号を分離・抽出することで、マイクロフォン数が多い場合でも、独立成分分析に関する演算量を低減することができ、演算時間の低減に寄与することができる。
【００７６】
なお、上述した第１及び第２の実施の形態では、複数のセンサとして、複数のアンテナ及び複数のマイクロフォンを用いたが本発明はこれらに限定されるものではない。複数のセンサとして、例えば、生体の種々の状態を検知するセンサを用いることができる。
【００７７】
なお、独立成分分析部５０におけるブラインド信号分離、独立成分分析などの処理内容については、上述した内容と同様であるので、その説明を省略する。
【発明の効果】
以上詳述したように、本発明によれば、複数の到来波を各々に入射するためのセンサ数が多い場合でも、独立成分分析に関する演算量を低減することができ、演算時間の低減に寄与することができる。
【図面の簡単な説明】
【図１】本発明の第１の実施の形態に係る混合信号分離・抽出装置の構成を示すブロック図である。
【図２】本発明の第２の実施の形態に係る混合信号分離・抽出装置の構成を示すブロック図である。
【符号の説明】
１１_１〜１１_ｎ　第１〜第ｎアンテナ
２１_１〜２１_ｎ　第１〜第ｎ帯域制限ろ波器
２５　帯域制限制御器
３０、９０　サンプリング部
３１〜３４　第１〜第４中間周波数変換器
３５　局部発振器
４１_１〜４１_ｎ　第１〜第ｎＡ／Ｄ変換器
４５　発振器
４７　到来波推定処理部
５０　独立成分分析部
６０　出力処理部
６１_１〜６１_ｎ　第１〜第ｎＤ／Ａ変換器
７１_１〜７１_ｎ　第１〜第ｎマイクロフォン
８１_１〜８１_ｎ　第１〜第ｎ帯域制限ろ波モジュール
９１_１〜９１_ｎ　第１〜第ｎ増幅器[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a blind signal separation device in which a plurality of sensors are arranged, and particularly to a mixed signal separation / extraction device capable of reducing the amount of calculation even when the number of sensors is large.
[0002]
[Prior art]
Conventionally, as a method of obtaining an original signal from a mixed signal of a plurality of signals reaching each of a plurality of sensors arranged at a predetermined interval, a method of blind signal separation (BSS: Blind Source Separation) is known.
[0003]
Generally, when a received signal is separated and extracted by a blind signal separation technique from observation signals observed by a plurality of sensors, a calculation is performed using digital data obtained by sampling a mixed signal reaching each sensor at a certain time interval. .
[0004]
[Problems to be solved by the invention]
Generally, in a method of calculating blind signal separation, a matrix operation is performed.
However, when the number of sensors is large, when a matrix operation is performed using all data observed by each sensor, the number of data increases, the amount of operation increases, and as a result, the operation time increases.
[0005]
The present invention has been made in view of the above, and an object of the present invention is to provide a mixed signal separation / extraction device capable of reducing the amount of calculation even when the number of sensors is large.
[0006]
[Means for Solving the Problems]
In order to solve the above problem, the invention according to claim 1 includes a plurality of sensors to which a plurality of incoming waves are respectively incident, and a plurality of mixed signals obtained by mixing the plurality of incident signals from the plurality of sensors. A plurality of band-limited filters each of which is output to be limited to a predetermined narrow band, based on a plurality of narrow band-limited observation signals from the plurality of band-limited filters, the number of arriving wave signals An arriving wave estimation processor for estimating and selecting the estimated number of observation signals from the plurality of observation signals, and an independent component based on the estimated number of observation signals selected by the arriving wave estimation processor. The gist is that an independent component analysis unit that separates and extracts the estimated number of incoming wave signals by an analysis method is provided.
[0007]
According to the first aspect of the present invention, as shown in FIG. 1, a plurality of arriving waves are respectively incident on the first to n-th antennas 11.₁~ 11_nAre arranged at predetermined intervals, and the first to n-th antennas 11₁~ 11_nA plurality of mixed signals obtained by mixing a plurality of incident signals from the first through n-th band-limited filters 21₁~ 21_n, Respectively, and the arriving wave estimation processing unit 47 estimates the number of arriving wave signals based on the plurality of observation signals limited to the narrow band, and calculates the estimated number from the plurality of observation signals. In the independent component analysis unit 50, the estimated number of arriving signals are separated and extracted by the independent component analysis method based on the selected estimated number of observed signals, thereby obtaining the number of antennas. Even when there are many, the amount of calculation for the independent component analysis can be reduced, which can contribute to a reduction in calculation time.
[0008]
According to a second aspect of the present invention, in order to solve the above-described problem, the plurality of sensors include a plurality of antennas or microphones arranged at predetermined intervals.
[0009]
According to a third aspect of the present invention, in order to solve the above problem, a plurality of sensors to which a plurality of incoming waves are respectively incident are arranged at predetermined intervals, and the plurality of incident signals from the plurality of sensors are mixed. Each of the plurality of mixed signals is limited to a predetermined narrow band, and based on the plurality of observation signals restricted to the narrow band, the number of arriving wave signals is estimated and the estimated number of observation signals is calculated from the plurality of observation signals. The gist is that an observation signal is selected, and the estimated number of incoming wave signals is separated and extracted by an independent component analysis method based on the selected estimated number of observation signals.
[0010]
According to a fourth aspect of the present invention, in order to solve the above problem, the plurality of mixed signals from the plurality of sensors is a plurality of mixed signals output from a plurality of antennas or a plurality of microphones.
[0011]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
The mixed signal separating / extracting apparatus according to this embodiment has an incident signal S_{1, ..., n}Is obtained by a blind signal separation technique. Blind signal separation is an algorithm that separates incident signals up to the number of sensors without prior knowledge of the responsiveness of the sensors, the properties of the signals, and the incident signals. This algorithm is called blind signal separation or independent component analysis (ICA: Independent Component Analysis), and many known documents have been published.
[0012]
(First Embodiment)
The mixed signal separation / extraction device according to the first embodiment of the present invention uses an antenna as a sensor, and separates and outputs an original signal from a mixed signal of a plurality of incident signals incident on a plurality of antennas. In the following, for simplicity, the number of antennas is described as “n” and the maximum number of incident signals is described as “n”, but the number of antennas and the number of incident signals are not limited to these. Optional.
[0013]
FIG. 1 is a block diagram showing a configuration of a mixed signal separating / extracting apparatus according to the first embodiment of the present invention.
The mixed signal separating / extracting device includes first to n-th antennas 11₁~ 11_n, The first to n-th band-limited filters 21₁~ 21_n, A band limiting controller 25, a sampling unit 30, an incoming wave estimation processing unit 47, an independent component analysis unit 50, and an output processing unit 60.
[0014]
1st to nth antenna 11₁~ 11_nFor example, a non-directional antenna such as a vertical antenna and a dipole antenna, a Yagi antenna having an arbitrary directivity, and the like are used to receive radio waves from various directions. These first to n-th antennas 11₁~ 11_nThe intervals and heights at which they are installed are arbitrary.
[0015]
First antenna 11₁Represents a plurality of incident signals (radio waves) S from the air₁~ S_nAnd a mixed signal obtained by mixing them is converted to a first band-limited filter 21.₁Send to Similarly, the second to n-th antennas 11₂~ 11_nRepresents a plurality of incident signals S from the air₂~ S_nAnd a mixed signal obtained by mixing these is filtered by a second to n-th band-limited filter 21.₂~ 21_nTo each.
[0016]
First band-limited filter 21₁Is the first antenna 11₁And passes only the frequency components in a predetermined band included in the mixed signal from the sampling unit 30 to the sampling unit 30. Similarly, the second to n-th band-limited filters 21₂~ 21_nAre the second to n-th antennas 11₂~ 11_nAnd passes only the frequency components of a predetermined band included in the mixed signal from to the sampling unit 30. Note that each of the first to n-th band-limited filters 21₁~ 21_nPass the same frequency band.
[0017]
The band limitation controller 25 generates a control signal for designating a frequency band to be passed, that is, a range of frequency components to be passed. The control signal generated by the band-limiting controller 25 includes the first to n-th band-limited filters 21.₁~ 21_nSent to First to n-th band-limited filters 21₁~ 21_nAccording to the control signal from the band limiting controller 25, only the frequency component of a predetermined band included in the input mixed signal is passed.
[0018]
The band limitation controller 25 is configured to generate a control signal that allows a frequency component in an arbitrary band to pass. Therefore, by appropriately changing the control signal from the band-limiting controller 25, the first to n-th band-limited filters 21₁~ 21_nCan be arbitrarily changed.
[0019]
The sampling unit 30 includes first to n-th intermediate frequency converters 31.₁~ 31_n, Local oscillator 35, first to n-th A / D converters 41₁~ 41_nAnd an oscillator 45.
The local oscillator 35 generates a signal having a transmission frequency required to convert a received radio wave to an intermediate frequency. The signal generated by the local oscillator 35 is transmitted to the first to n-th intermediate frequency converters 31.₁~ 31_nSent to
First to n-th frequency converter 31₁~ 31_nAlthough not shown, each of them is composed of a high frequency amplifier, a frequency mixer and an intermediate frequency amplifier.
[0020]
The high frequency amplifier amplifies a high frequency in a reception frequency band so as to have an appropriate magnitude as an input voltage of the next-stage frequency mixer. The frequency mixer mixes the signal amplified by the high frequency amplifier and the output signal of the local oscillator 35, and converts the signal into an intermediate frequency signal by forming a sum or difference frequency. The intermediate frequency amplifier converts the frequency of the received radio wave to a lower intermediate frequency and amplifies it. Thereby, stable and high-gain amplification can be performed, and sensitivity can be improved.
The first to n-th intermediate frequency converters 31 configured as described above₁~ 31_nAre output from the first to n-th A / D converters 41.₁~ 41_nSent to each.
[0021]
The oscillator 45 includes the first to n-th intermediate frequency converters 31.₁~ 31_nGenerates a sampling clock for sampling the signal from the. The sampling clock generated by the oscillator 45 is supplied to the first to n-th A / D converters 41.₁~ 41_nSent to
[0022]
First to n-th A / D converters 41₁~ 41_nAre the first to n-th intermediate frequency converters 31 using the signal from the oscillator 45 as a sampling clock.₁~ 31_nBy converting the analog signals into digital signals. First to n-th A / D converters 41₁~ 41_nAre supplied to the arriving wave estimation processing unit 47.
[0023]
The arriving wave estimation processing unit 47 estimates the number k of signals of the arriving wave and calculates n sensor signals X_{1 , Λ , N}(T) is selected at random from k observation signals x_{1, ..., k}The result is sent to the independent component analyzer 50 as (t). The details of the incoming wave estimation processing unit 47 will be described later. Observation signal x output from incoming wave estimation processing section 47_{1, ..., k}(T) is supplied to the independent component analyzer 50.
[0024]
The independent component analysis unit 50 determines the original signal (the incident signal S₁~ S₄) Is separated and extracted from the observation signal obtained by mixing the above. Details of the independent component analyzer 50 will be described later. The signal extracted by the independent component analyzer 50 is sent to the output processor 60.
[0025]
The output processing unit 60 includes the first to k-th D / A converters 61.₁~ 61_kIt is composed of First to k-th D / A converter 61₁~ 61_kConverts the digital signal from the independent component analyzer 50 into an analog signal, and outputs the separated signal O₁~ O_kRespectively.
[0026]
Next, the operation of the mixed signal separating / extracting apparatus according to the first embodiment of the present invention configured as described above will be described.
[0027]
Each antenna 11₁~ 11_nSignal S observed at_{1, ..., n}Mixed signal x_{1, ..., n}(T) shows the first to n-th band-limited filters 21₁~ 21_nSent to each. First to n-th band-limited filters 21₁~ 21_nIs the mixed signal x_{1, ..., n}Only the frequency components in the predetermined band of (t) are passed, and sent to the sampling unit 30.
[0028]
First intermediate frequency converter 31 of sampling section 30₁Is the first band-limited filter 21₁Mixed signal x from₁(T) is input and converted into an intermediate frequency signal.
Specifically, the first intermediate frequency converter 31₁The high frequency amplifier inside is a mixed signal x₁(T) is amplified by high frequency. The high-frequency amplified signal is mixed with a signal from the local oscillator 35 to generate a sum or difference frequency between them, thereby being converted into an intermediate frequency signal. The signal of the intermediate frequency is amplified by the intermediate frequency amplifier and the first A / D converter 41₁Sent to
[0029]
First A / D converter 41₁Is the first intermediate frequency converter 31₁Is converted into a digital signal by sampling using the sampling clock from the oscillator 45, and the observation signal X₁Output as (t).
[0030]
Similarly, the second to n-th intermediate frequency converters 31₂~ 31_nAre the second to n-th band-limited filters 21₂~ 21_nMixed signal x from_{2, ..., n}(T) is input to each of them and converted into an intermediate frequency signal.₂~ 41_nTo each.
[0031]
Second to nth A / D converters 41₂~ 41_n, The second to n-th intermediate frequency converters 31₂~ 31_nIs converted into a digital signal by sampling using the sampling clock from the oscillator 45, and the observation signal X_{2, ..., n}Output as (t). At this time, the sampling interval (frequency of the sampling clock) is₁~ 11_nSignal S arriving at_{1, ..., n}Is adjusted to a value that does not affect the time difference. The observation signal X generated by the sampling unit 30 in this manner_{1, ..., n}(T) is sent to the incoming wave estimation processing unit 47.
[0032]
The arriving wave estimation processing unit 47 includes the first to n-th A / D converters 41.₁~ 41_nObservation signal X from_{1, ..., n}(T) is input, and the estimated number k of incoming wave signals is estimated based on first and second methods described later.
[0033]
Here, in the first method, the observed time-series data X_{1 , Λ , N}The correlation matrix of (t) is R_xxExpressed as the eigenvalue of
(Equation 1)

Is obtained.
[0034]
First, the eigenvalue of the correlation matrix of the observation data is obtained from this equation (1), and the noise power σ²The estimated number m of incoming wave signals can be estimated from the number of eigenvalues for the larger, substantially valid signal.
[0035]
Next, in the second method, for example, the literature (MATI @ MAX, THOMAS @ KAILATH, "Detection of Signals by Information" Theoretic Criteria "," IEEE Transactions, Agency, Sponsor, Acoustics, Acronym, Acoustics, Acoustics, Acoustics, Acronym, Acoustics, Acronym, Acoustics, Acronym, Acoustics, Acronym, Acronym, Acronym, Acronym, Acronym, Acronym, Acronym, Acronym, Acronym, Acronym, Acronym) ) Include AIC (Akaike Information Criteria) and MDL (Minimum Description Length).
[0036]
AIC (k) is in the range of k = 1 to n,
(Equation 2)

Is calculated, and the minimum value of k is estimated as the estimated number k of the incoming wave signal.
[0037]
MDL (k) is in the range of k = 1 to n,
(Equation 3)

Is calculated, and the minimum value of k is estimated as the estimated number k of the incoming wave signal.
[0038]
As described above, the arriving wave estimation processing unit 47 estimates the estimated numbers m and k of the arriving wave signals using the first and second methods described above, and sets the estimated numbers m and k to (m ≦ k). <N), X_{1 , Λ , N}K sensor signals X that satisfy (m ≦ k <n) from (t)_{1 , Λ , N}(T) is randomly selected and the observation signal x_{1, ..., k}The result is sent to the independent component analyzer 50 as (t).
[0039]
Also, when the estimated number k of arriving wave signals received by a plurality of antennas can be estimated, the blind signal separation is performed using signals of the number of sensors that is larger than the estimated number k and smaller than the number of all antennas. By doing so, the amount of calculation may be reduced.
[0040]
Note that X_{1 , Λ , N}As a method of randomly selecting k items from (t), for example, the first to k-th X_{1 , Λ , k}(T) is selected in order of k numbers, k non-overlapping random numbers in the range of 1 to n are generated, and this random number X_{1 , Λ , k}There is a method of selecting only k pieces of (t).
[0041]
The independent component analyzer 50 calculates the observation signal x_{1, ..., k}Using the k sampling time series transmitted as (t), a signal y (t) in the selected band is separated and extracted by a blind signal separation algorithm.
[0042]
Hereinafter, the procedure of separation and extraction will be described in detail.
The blind signal separation in the first embodiment is used for separating signals mixed without time delay. First, the fundamental problem of blind signal separation is defined here. It is assumed that the signal source is given by the vector of the following equation (4).
[0043]
(Equation 4)

Here, it is assumed that s (t) is n incident signals, has an average of “0”, and is independent of each other. T represents transposition.
[0044]
The observation signal is transmitted to each antenna 11₁~ 11_nMeans time-series data observed and band-limited,
(Equation 5)

It shall be represented by This means that each antenna 11 of 1,...₁~ 11_nCan be considered as the signal observed at Generally, the number of antennas does not always match the number of signal sources, but it is assumed here that they match.
[0045]
In a simple ICA problem, between s (t) and x (t),
(Equation 6)

Assume a simple linear relationship A indicates each antenna 11₁~ 11_nIs a real number matrix of a signal mixing matrix (n rows × n columns) determined by the arrangement and characteristics of. x (t) is separated into independent signal components without knowledge of s (t) or A.
[0046]
That is, by obtaining a certain n × n real number matrix,
(Equation 7)

The purpose of ICA is to reconstruct mutually independent y (t) determined by B is ideally A^-1However, this is not the case, and the arbitrariness of permutation and amplitude remains.
[0047]
One of the solutions to this problem is a separation method based on the independence of probability distributions. Each s_iUnder the assumption that (t) is (strong) stationary and not Gaussian, y_iVarious methods have been proposed for obtaining B so that (t) is independent of each other, and many of them can be summarized as follows. Assuming that y (t) is a strongly stationary process, the density function of the simultaneous distribution is
(Equation 8)

Then the definition of independence is p (y_i) To p (y) y_iAs a marginal distribution on,
(Equation 9)

And multiply.
[0048]
The Kullback-Leibler divergence between the product of the joint distribution and the marginal distribution is
(Equation 10)

Becomes Here, H (Y; B) is the entropy of the joint distribution p (y), and H (Y_iB) is a marginal distribution p (y_i) Entropy.
[0049]
This is ｛Y_i相互 (i = 1,..., N). Based on the assumption that the signal source is not normally distributed, KL (B) becomes p (y_i) Is “0” only when they are independent of each other. These are determined by p (x) and B.
[0050]
Note that p (y) dy = p (x) dx and p (y) = p (x) / | B | (| B | is the determinant of B).
[Equation 11]

Becomes
[0051]
On the other hand, the entropy of the marginal distribution is
(Equation 12)

It is. Therefore,
(Equation 13)

Becomes
[Equation 14]

Then, the correct B can be obtained as the steepest descent method. The problem in equation (14) is that the inverse matrix (B^T)^-1Is calculated.
[0052]
Regarding the convergence, since this may be multiplied by any positive definite matrix, B^TMultiplying by B (which corresponds to a Rieman metric on a manifold of a regular matrix)
[Equation 15]

Becomes a new learning rule. By the assumption of stationarity, p (s), p (x) and p (y) are temporally independent. Under this assumption, the ensemble average can be replaced by a time average.
[0053]
(Equation 16)

Therefore, if η is a positive constant and the parameter is updated according to the following equation (17) every time data is observed, B_tIs obtained.
[0054]
[Equation 17]

Here, the problem naturally arises in p (y) in the above equation (13)._i) Or ψ (y). Usually, this uses a parametric nonlinear function or a statistical expansion method.
[0055]
A rough idea is that if p (y_i) Is a normal distribution, then ψ (y) is a linear function. On the other hand, when the tail is "heavier" than the normal distribution (sub-Gaussian), it is better to approximate with a polynomial, and when the tail is "lighter" than the normal distribution (super-Gaussian), it is better to approximate with a sigmoid function. It is good. The sigmoid function works well because the tail of the audio signal is "light".
The signal y separated and extracted by the independent component analyzer 50 as described above_{1, ..., k}(T) is sent to the output processing unit 60.
[0056]
In the output processing unit 60, the signal y sent as a digital signal from the independent component analysis unit 50_{1, ..., k}(T) is converted to an analog signal. More specifically, the signal y output from the output processing unit 60₁(T) is converted into a digital signal by the first D / A converter 61 and the separated signal O₁Is sent to the outside. Similarly, the signal y_{2, ..., k}(T) is the second to k-th D / A converter 61₂~ 61_kAre converted into digital signals respectively, and the separated signal O₂~ O_kAre sent to the outside.
[0057]
As a result, since the independent component analysis unit 50 performs the independent component analysis operation using the k observation signals, the amount of operation can be reduced as compared with the conventional operation using n units, which contributes to shortening the operation time. it can.
[0058]
As described above, according to the mixed signal separation / extraction device according to the first embodiment, the first to n-th antennas 11 into which a plurality of arriving waves are incident respectively.₁~ 11_nAre arranged at predetermined intervals, and the first to n-th antennas 11₁~ 11_nA plurality of mixed signals obtained by mixing a plurality of incident signals from the first through n-th band-limited filters 21₁~ 21_n, Respectively, and the arriving wave estimation processing unit 47 estimates the number of arriving wave signals based on the plurality of observation signals limited to the narrow band, and calculates the estimated number from the plurality of observation signals. In the independent component analysis unit 50, the estimated number of arriving signals are separated and extracted by the independent component analysis method based on the selected estimated number of observed signals, thereby obtaining the number of antennas. Even when there are many, the amount of calculation for the independent component analysis can be reduced, which can contribute to a reduction in calculation time.
[0059]
(Second embodiment)
The mixed signal separating / extracting apparatus according to the second embodiment of the present invention uses a microphone as a sensor, and separates and outputs an original signal from a mixed signal of a plurality of incident signals incident on a plurality of microphones. In the following, for the sake of simplicity, the number of antennas is assumed to be “n” and the number of incident signals is assumed to be “n”. It is. Further, the same components as those in the first embodiment are denoted by the same reference numerals, and description thereof will be omitted or simplified.
[0060]
FIG. 2 is a block diagram showing a configuration of a mixed signal separating / extracting apparatus according to the second embodiment of the present invention.
The mixed signal separating / extracting device includes first to n-th microphones 71₁~ 71_n, First to n-th band-limited filtering module 81₁~ 81_n, A band limiting controller 25, a sampling unit 90, an incoming wave estimation processing unit 47, an independent component analysis unit 50, and an output processing unit 60.
[0061]
First to n-th microphones 71₁~ 71_nFor example, various microphones having omnidirectionality and arbitrary directivity can be used, and receive sound from all directions. These first to n-th microphones 71₁~ 71_nThe intervals and heights at which they are installed are arbitrary.
[0062]
First microphone 71₁Represents a plurality of incident signals (sound waves) S from the air₁~_nAnd a mixed signal obtained by mixing them is converted to a first band-limited filter module 81.₁Send to Similarly, the second to n-th microphones 71₂~ 71_nRepresents a plurality of incident signals S from the air₁~ S_nAnd a mixed signal obtained by mixing these signals is supplied to the second to n-th band-limited filtering modules 81.₁~ 81_nTo each.
[0063]
First band limited filtering module 81₁Is the first microphone 71₁It comprises a preamplifier for amplifying a mixed signal from the CDMA and a band-limiting filter for passing only frequency components of a predetermined band included in the mixed signal. This first band limited filtering module 81₁Is the first microphone 71₁, And passes only frequency components of a predetermined band included in the amplified mixed signal to the sampling unit 90.
[0064]
Similarly, the second to n-th band-limited filtering modules 81₂~ 81_nAre the second to n-th microphones 71₂~ 71_nIt comprises a preamplifier for amplifying a mixed signal from the CDMA and a band-limiting filter for passing only frequency components of a predetermined band included in the mixed signal. These second to fourth band-limited filtering modules 81₂~ 81_nAre the second to fourth microphones 71₂~ 71_nAre amplified, and only the frequency components of a predetermined band included in the amplified mixed signal are passed and sent to the sampling unit 90. Note that each of the first to n-th band-limited filtering modules 81₁~ 81_nPass the same frequency band.
[0065]
The band limit controller 25 is the same as that of the first embodiment. Therefore, by appropriately changing the control signal from the band-limiting controller 25, the first to fourth band-limited filtering modules 81₁~ 81_nCan be arbitrarily changed.
[0066]
The sampling unit 90 includes first to n-th amplifiers 91.₁~ 91_n, The first to n-th A / D converters 41₁~ 41_nAnd an oscillator 45.
First to n-th amplifiers 91₁~ 91_nAre the first to n-th band-limited filtering modules 81₁~ 81_nFrom each other. First to n-th amplifiers 91₁~ 91_nAre amplified by the first to n-th A / D converters 41₁~ 41_nSent to each. First to n-th A / D converters 41₁~ 41_nThe configuration of the oscillator 45 is the same as that of the first embodiment. First to n-th A / D converters 41₁~ 41_nAre supplied to the arriving wave estimation processing unit 47.
The configurations of the incoming wave estimation processing unit 47, the independent component analysis unit 50, and the output processing unit 60 are the same as those of the first embodiment.
[0067]
Next, the operation of the mixed signal separating / extracting apparatus according to the second embodiment of the present invention configured as described above will be described.
Each microphone 71₁~ 71_nIncident signals S observed at_{1, ..., n}Mixed signal x_{1, ..., n}(T) shows the first to n-th band-limited filtering modules 81₁~ 81_nSent to each. First to n-th band-limited filtering module 81₁~ 81_nIs the mixed signal x_{1, ..., n}After amplifying (t), only the frequency components in the predetermined band are passed, and sent to the sampling unit 90.
[0068]
First amplifier 91 of sampling section 90₁Is a first band limited filtering module 81₁Mixed signal x from₁(T) is input and amplified. The signal amplified by the first amplifier 91 is supplied to the first A / D converter 41.₁Sent to
[0069]
First A / D converter 41₁Is the first amplifier 91₁Is converted into a digital signal by sampling using the sampling clock from the oscillator 45, and the observation signal X₁Output as (t).
[0070]
Similarly, the second to n-th amplifiers 91₁~ 91_nAre the second to n-th band-limited filtering modules 81₂~ 81_nMixed signal x from_{2, ..., n}(T) are inputted and amplified, and the second to n-th A / D converters 41 are inputted.₂~ 41_nTo each.
[0071]
Second to nth A / D converters 41₂~ 41_nAre the second to n-th amplifiers 91₂~ 91_nIs converted into a digital signal by sampling using the sampling clock from the oscillator 45, and the observation signal X_{2, ..., n}Output as (t). At this time, the sampling interval (frequency of the sampling clock) is₂~ 71_nSignal S arriving at_{1, ..., n}Is adjusted to a value that does not affect the time difference. Observation signal X generated by sampling section 90 in this manner_{1, ..., n}(T) is sent to the incoming wave estimation processing unit 47.
[0072]
The arriving wave estimation processing unit 47 estimates the number of signals m and k of the arriving wave using the above-described first and second techniques, and when (m ≦ k <n) is satisfied, X_{1 , Λ , N}K sensor signals X that satisfy (m ≦ k <n) from (t)_{1 , Λ , N}(T) is randomly selected and the observation signal x_{1, ..., k}The result is sent to the independent component analyzer 50 as (t).
[0073]
The independent component analyzer 50 calculates the observation signal x_{1, ..., n}Using the k sampling time series transmitted as (t), a signal y (t) in the selected band is separated and extracted by a blind signal separation algorithm. The procedure of separation / extraction is the same as the procedure described in the first embodiment. The signal y separated and extracted by the independent component analyzer 50_{1, ..., n}(T) is sent to the output processing unit 60.
[0074]
In the output processing unit 60, the signal y sent as a digital signal from the independent component analysis unit 50_{1, ..., k}(T) is converted to an analog signal. More specifically, the signal y output from the output processing unit 60₁(T) is converted into a digital signal by the first D / A converter 61 and the separated signal O₁Is sent to the outside. Similarly, the signal y_{2, ..., k}(T) is the second to n-th D / A converter 61₁~ 61_kAre converted into digital signals respectively, and the separated signal O₂~ O_kAre sent to the outside.
[0075]
As described above, according to the mixed signal separating / extracting apparatus according to the second embodiment, the microphone 71 into which a plurality of incoming waves are incident respectively.₁~ 71_nAre arranged at predetermined intervals, and the first to n-th microphones 71₁~ 71_nA plurality of mixed signals obtained by mixing a plurality of incident signals from the first to n-th band-limited filtering modules 81₁~ 81_n, Respectively, and the arriving wave estimation processing unit 47 estimates the number of arriving wave signals based on the plurality of observation signals limited to the narrow band, and calculates the estimated number from among the plurality of observation signals. The independent component analysis unit 50 separates and extracts the estimated number of arriving wave signals by the independent component analysis method based on the selected estimated number of observed signals, thereby obtaining the number of microphones. Even when there are many, the amount of calculation for the independent component analysis can be reduced, which can contribute to a reduction in calculation time.
[0076]
In the first and second embodiments described above, a plurality of antennas and a plurality of microphones are used as the plurality of sensors, but the present invention is not limited to these. As the plurality of sensors, for example, sensors that detect various states of a living body can be used.
[0077]
The details of processing such as blind signal separation and independent component analysis in the independent component analysis unit 50 are the same as those described above, and a description thereof will be omitted.
【The invention's effect】
As described above in detail, according to the present invention, even when the number of sensors for inputting a plurality of arriving waves is large, it is possible to reduce the amount of calculation related to the independent component analysis, which contributes to the reduction of the calculation time. can do.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of a mixed signal separation / extraction device according to a first embodiment of the present invention.
FIG. 2 is a block diagram illustrating a configuration of a mixed signal separation / extraction device according to a second embodiment of the present invention.
[Explanation of symbols]
11₁~ 11_n1st to nth antenna
21₁~ 21_nFirst to nth band-limited filters
25 band limiting controller
30, 90 ° sampling unit
31-34 first to fourth intermediate frequency converters
35 ° local oscillator
41₁~ 41_nFirst to nth A / D converters
45 ° oscillator
47 Arrival wave estimation processing unit
50 independent component analysis unit
60 ° output processing unit
61₁~ 61_nFirst to nth D / A converters
71₁~ 71_nFirst to nth microphones
81₁~ 81_nFirst to nth band-limited filtering modules
91₁~ 91_nFirst to n-th amplifiers

Claims (4)

A plurality of sensors to each of which a plurality of incoming waves are incident,
A plurality of band-limiting filters that output a plurality of mixed signals obtained by mixing the plurality of incident signals from the plurality of sensors, each of which is limited to a predetermined narrow band,
Based on a plurality of observation signals restricted to a narrow band from the plurality of band-limited filters, the number of arriving wave signals is estimated, and the estimated number of observation signals is selected from the plurality of observation signals. An incoming wave estimation processing unit;
An independent component analysis unit that separates and extracts the estimated number of incoming wave signals by an independent component analysis method based on the estimated number of observed signals selected by the incoming wave estimation processing unit. Characteristic mixed signal separation / extraction device. The plurality of sensors,
The mixed signal separation / extraction device according to claim 1, comprising a plurality of antennas or a plurality of microphones arranged at predetermined intervals. A plurality of sensors to which a plurality of incoming waves are respectively incident are arranged at predetermined intervals,
Each of the plurality of mixed signals obtained by mixing the plurality of incident signals from the plurality of sensors is limited to a predetermined narrow band,
Based on the plurality of observation signals restricted to the narrow band, from the plurality of observation signals to estimate the number of arriving wave signals, select the observation signal of the estimated number from among the plurality of observation signals,
A mixed signal separating / extracting method, wherein the estimated number of incoming wave signals is separated / extracted by an independent component analysis technique based on the selected estimated number of observed signals. The method according to claim 3, wherein the plurality of mixed signals from the plurality of sensors are a plurality of mixed signals output from a plurality of antennas or a plurality of microphones.

Images