JP4201907B2 - Signal separation device, signal separation method, and recording medium - Google Patents

Description

【００１０】
フィルタ処理手段７は伝送路逆特性演算手段９で算出されたタップ係数C(n)を元に伝送路フィルタ手段６からの出力信号X(n)に対してフィルタリング処理を行う（ステップ４）。
【００１１】
以上の工程を規定回数行うことにより伝送路逆特性値C(n)を学習させ、規定回数回トレーニング信号の受信が終了したならば（ステップ５）、切り替え手段３が原信号発生手段からの信号を加算手段５に出力する（ステップ６）。ステップ６で加算手段５に出力された信号は雑音発生手段４からの雑音信号と加算手段５で加算され、更に伝送路フィルタ手段６で減衰・歪みを与えられる。フィルタ処理手段７は伝送路フィルタ手段６からの出力信号に対してトレーニング信号受信時に学習した伝送路逆特性C(n)を用いてフィルタリング処理を行い（ステップ７）、伝送路フィルタ手段からの出力信号から雑音を取り除いた原信号に近い信号Y(n)を出力し、終了する（ステップ８）。
【００１２】
また、ステップ６、ステップ７、ステップ８においては伝送路逆特性値C(n)の演算は行わない。
【００１３】
【発明が解決しようとする課題】
しかしながら上記の構成では、原信号による通信以外にトレーニング用の信号を送受信する必要がありスループットが低下する。また、ステップ６、ステップ７、ステップ８においては伝送路逆特性値C(n)は算出しない為、伝送路の特性が原信号送受信時に時間的に変化する場合については対応できないという問題点を有していた。
【００１４】
【課題を解決するための手段】
上記課題を解決するために本発明は、送信側から受信側へ信号を伝達する場合伝達すべき所望信号（原信号）に対して、伝送路内で所望信号の識別に障害を来す雑音信号が任意の増幅率で重畳され、受信側において所望信号を識別できない場合において、受信側で雑音信号にうもれた所望信号を再生する為に、２入力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段と、フーリエ変換手段からの出力値に対して相互相関演算を行う相互相関演算手段と、フーリエ変換手段からの出力値に対して自己相関演算を行う自己相関演算手段と、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段と、相互相関演算手段とNeglect手段の出力値から伝送路のパラメータを推定するパラメータ推定手段を有しことにより、トレーニング信号を必要とせずに原信号と雑音を分離できる作用を有し、原信号送受信時に伝送路が時間的に変化する場合においても伝送路変化に追従して信号分離処理を行うことのできるアルゴリズムが提供できる。
【００１５】
【発明の実施の形態】
本発明の請求項１に記載の発明は、送信側から受信側へ信号を伝達する場合、伝達すべき所望信号（原信号）に対して、伝送路内で所望信号の識別に障害を来す雑音信号が任意の増幅率で重畳され、受信側において所望信号を識別できない場合において、受信側で雑音信号にうもれた所望信号を再生する為に、原信号発生手段から出力された原信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行う第１及び第２のフィルタ手段と、雑音発生手段から出力された雑音信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行う第３及び第４のフィルタ手段と、第１のフィルタ手段の出力と第３のフィルタ手段の出力を加算し、フィルタリングされた原信号と雑音を重ねあわせる第１の加算手段と、第２のフィルタ手段と第４のフィルタ手段の出力を加算し、フィルタリングされた雑音と原信号を重ねあわせる第２の加算手段と、第１の加算手段からの出力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段と、フーリエ変換手段からの出力値に対して相互相関演算を行う相互相関演算手段と、フーリエ変換手段からの出力値に対して自己相関演算を行う自己相関演算手段と、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段と、相互相関演算手段とNeglect手段の出力値から第１乃至第４のフィルタ手段及び第１、第２の加算手段を含む伝送路のパラメータを推定するパラメータ推定手段と、パラメータ推定手段で推定されたパラメータを元に伝送路の第１の加算手段からの出力値に対してフィルタリング処理を行う第５のフィルタ手段と、同じくパラメータ推定手段で推定されたパラメータを元に伝送路の第２の加算手段からの出力値に対してフィルタリング処理を行う第６のフィルタ手段と、伝送路の第１の加算手段からの出力値と第６のフィルタ手段の出力値を加算する第３の加算手段と、伝送路の第２の加算手段からの出力値と第５のフィルタ手段の出力値を加算する第４の加算手段と、第３の加算手段からの出力値に対して伝送路のパラメータを元に乗算処理を行う第１の乗算手段と、第４の加算手段からの出力値に対して伝送路のパラメータを元に乗算処理を行う第２の乗算手段を有する信号分離装置であり、トレーニング信号を必要とせずに伝送路からの出力値に対して原信号と雑音を分離させる信号分離処理を行い原信号に近い信号を抽出する作用がある。
【００１６】
本発明の請求項２に記載の発明は、ツイストペアケーブルを用いて送信側から受信側へ信号を伝達する場合、ツイストペアケーブル内にて接地線に信号線の信号が任意の増幅率で重畳され送信側と受信側の接地電圧値に揺らぎが生じ、結果として受信側において所望信号を識別できない場合において、受信側で送信側と同じ接地電圧値を得る為に、ツイストペアケーブルによる通信において、送信側接地電圧値としての原信号発生手段からのデジタル出力信号をアナログ信号へ変換するデジタル／アナログ変換手段と、デジタル／アナログ変換手段から出力された信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行う第１及び第２のフィルタ手段と、送信側信号線としての雑音発生手段から出力された雑音信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行う第３及び第４のフィルタ手段と、第１のフィルタ手段の出力と第３のフィルタ手段の出力を加算し、フィルタリングされたアナログ原信号と雑音を重ねあわせる第１の加算手段と、第２のフィルタ手段と第４のフィルタ手段の出力を加算し、フィルタリングされた雑音とアナログ原信号を重ねあわせる第２の加算手段と、第１の加算手段からのアナログ出力信号をデジタル信号へ変換する第１のアナログ／デジタル変換手段と、第２の加算手段からのアナログ出力信号をデジタル信号へ変換する第２のアナログ／デジタル変換手段と、第１及び第２のアナログ／デジタル変換手段からの出力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段と、フーリエ変換手段からの出力値に対して相互相関演算を行う相互相関演算手段と、フーリエ変換手段からの出力値に対して自己相関演算を行う自己相関演算手段と、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段と、相互相関演算手段とNeglect手段の出力値から第１乃至第４のフィルタ手段及び第１、第２の加算手段を含む伝送路のパラメータを推定するパラメータ推定手段と、パラメータ推定手段で推定されたパラメータを元にツイストペアケーブルの第１の加算手段からの出力値に対してフィルタリング処理を行う第５のフィルタ手段と、同じくパラメータ推定手段で推定されたパラメータを元にツイストペアケーブルの第２の加算手段からの出力値に対してフィルタリング処理を行う第６のフィルタ手段と、ツイストペアケーブルの加算手段からの出力値と第６のフィルタ手段の出力値を加算する第３の加算手段と、ツイストペアケーブルの第２の加算手段からの出力値と第５のフィルタ手段の出力値を加算する第４の加算手段と、第３の加算手段からの出力値に対してツイストペアケーブルのパラメータを元に乗算処理を行う第１の乗算手段と、第４の加算手段からの出力値に対してツイストペアケーブルのパラメータを元に乗算処理を行う第２の乗算手段を有する信号分離装置であり、ツイストペアケーブルによる通信においてトレーニング信号を必要とせずにツイストペアケーブルからの出力値に対して原信号と雑音を分離させる信号分離処理を行い原信号に近い信号を抽出し、受信側にて接地電圧値から雑音成分を取り除き、送信側と等しくでき安定した信号伝達を実現できるという作用がある。
【００１７】
本発明の請求項３に記載の発明は、カーナビゲーションシステムの音声認識装置や携帯型電話機、ＯＦＤＭ（Orthogonal Frequency Division Multiplexing :直交周波数分割多重方式）通信等において、カーナビゲーションシステムの音声認識装置では話者（送信側）からカーナビゲーション装置（受信側）へ音声コマンドを伝達する場合、音声コマンド（原信号）に対して、車内のエンジン装置やタイヤ装置の路面との摩擦音等の音声コマンドの識別に障害を来す雑音信号が任意の増幅率で重畳され、カーナビゲーション装置において音声コマンドを識別できない場合において、カーナビゲーション装置で雑音に埋もれた音声コマンドを再生する為に、また、携帯電話装置では話者（原信号）の音声信号に対して話者以外の周囲雑音が任意の増幅率で重畳され、話者の通話者では音声を認識できない場合において、送信側の携帯電話装置において雑音に埋もれた話者の声を再生する為に、また、ＯＦＤＭ通信においては複数の送信局からの信号を受信局にて所望局の信号（原信号）のみを取り出す為に、雑音（ＯＦＤＭ通信では他チャンネルの信号）が混入した所望信号から所望信号のみを抽出する場合において、原信号発生手段から出力された信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行う第１及び第２のフィルタ手段と、雑音発生手段から出力された雑音信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行う第３及び第４のフィルタ手段と、第１のフィルタ手段の出力と第３のフィルタ手段の出力を加算し、フィルタリングされた原信号と雑音を重ねあわせる第１の加算手段と、第２のフィルタ手段と第４のフィルタ手段の出力を加算し、フィルタリングされた雑音と原信号を重ねあわせる第２の加算手段と、第１の加算手段からの出力信号を電気信号へと変換する第１のマイクロフォンと、第２の加算手段からの出力信号を電気信号へと変換する第２のマイクロフォンと、第１のマイクロフォンからのアナログ出力信号をデジタル信号へ変換する第１のアナログ／デジタル変換手段と、第２のマイクロフォンからのアナログ出力信号をデジタル信号へ変換する第２のアナログ／デジタル変換手段と、第１及び第２のアナログ／デジタル変換手段からの出力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段と、フーリエ変換手段からの出力値に対して相互相関演算を行う相互相関演算手段と、フーリエ変換手段からの出力値に対して自己相関演算を行う自己相関演算手段と、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段と、相互相関演算手段とNeglect手段の出力値から第１乃至第４のフィルタ手段及び第１、第２の加算手段を含む伝送路のパラメータを推定するパラメータ推定手段と、パラメータ推定手段で推定されたパラメータを元に伝送路の第１の加算手段からの出力値に対してフィルタリング処理を行う第５のフィルタ手段と、同じくパラメータ推定手段で推定されたパラメータを元に伝送路の第２の加算手段からの出力値に対してフィルタリング処理を行う第６のフィルタ手段と、伝送路の第１の加算手段からの出力値と第６のフィルタ手段の出力値を加算する第３の加算手段と、伝送路の第２の加算手段からの出力値と第５のフィルタ手段の出力値を加算する第４の加算手段と、第３の加算手段からの出力値に対して伝送路のパラメータを元に乗算処理を行う第１の乗算手段と、第４の加算手段からの出力値に対して伝送路のパラメータを元に乗算処理を行う第２の乗算手段を有する信号分離装置であり、トレーニング信号を必要とせずに伝送路からの出力値に対して原信号と雑音を分離させる信号分離処理を行い原信号に近い信号を抽出する作用がある。
【００１８】
本発明の請求項４に記載の発明は、２入力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段と、フーリエ変換手段からの出力値に対して相互相関演算を行う相互相関演算手段と、フーリエ変換手段からの出力値に対して自己相関演算を行う自己相関演算手段と、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段と、相互相関演算手段とNeglect手段の出力値から伝送路のパラメータを推定するパラメータ推定手段を有し、相互相関演算手段と自己相関演算手段を並列処理することにより高速演算を行うことを特徴とする信号分離装置であり、演算に時間のかかる自己相関演算と相互相関演算を並列処理することにより高速演算性、リアルタイム演算性を実現できる作用がある。
【００１９】
本発明の請求項５に記載の発明は、送信側から受信側へ信号を伝達する場合、伝達すべき所望信号（原信号）に対して、伝送路内で所望信号の識別に障害を来す雑音信号が任意の増幅率で重畳され、受信側において所望信号を識別できない場合において、受信側で雑音信号にうもれた所望信号を再生する為に、原信号発生手段から出力された原信号に対して第１及び第２のフィルタ手段にて任意タップ長かつ任意タップ係数のフィルタリング処理を行うステップと、雑音発生手段から出力された雑音信号に対して第３及び第４のフィルタ手段にて任意タップ長かつ任意タップ係数のフィルタリング処理を行うステップと、第１のフィルタ手段の出力と第３のフィルタ手段の出力を第１の加算手段にて加算し、フィルタリングされた原信号と雑音を重ねあわせるステップと、第２のフィルタ手段と第４のフィルタ手段の出力を第２の加算手段にて加算し、フィルタリングされた雑音と原信号を重ねあわせるステップと、第１の加算手段及び第２の加算手段からの出力信号に対しフーリエ変換手段にて短時間フーリエ変換処理もしくはフーリエ変換処理を行うステップと、フーリエ変換手段からの出力値に対し相互相関演算手段にて相互相関演算を行うステップと、フーリエ変換手段からの出力値に対し自己相関演算手段にて自己相関演算を行うステップと、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値についてはNeglect手段にて無視するステップと、相互相関演算手段とNeglect手段の出力値から第１乃至第４のフィルタ手段及び第１、第２の加算手段を含む伝送路のパラメータをパラメータ推定手段にて推定するステップと、パラメータ推定手段で推定されたパラメータを元に伝送路の第１の加算手段からの出力値に対して第５のフィルタ手段にてフィルタリング処理を行うステップと、同じくパラメータ推定手段で推定されたパラメータを元に伝送路の第２の加算手段からの出力値に対して第６のフィルタ手段にてフィルタリング処理を行うステップと、伝送路の第１の加算手段からの出力値と第６のフィルタ手段の出力値を第３の加算手段にて加算するステップと、伝送路の第２の加算手段からの出力値と第５のフィルタ手段の出力値を第３の加算手段にて加算するステップと、第３の加算手段からの出力値に対して伝送路のパラメータを元に第１の乗算手段にて乗算処理を行うステップと、第４の加算手段からの出力値に対して伝送路のパラメータを元に第２の乗算手段にて乗算処理を行うステップとを有する信号分離方法であり、トレーニング信号を必要とせずに伝送路からの出力値に対して原信号と雑音を分離させる信号分離処理を行い原信号に近い信号を抽出する作用がある。
【００２０】
本発明の請求項６に記載の発明は、２入力信号に対してフーリエ変換手段にて短時間フーリエ変換処理もしくはフーリエ変換処理を行うステップと、フーリエ変換手段からの出力値に対して相互相関演算手段にて相互相関演算を行うステップと、フーリエ変換手段からの出力値に対して自己相関演算手段にて自己相関演算を行うステップと、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値についてはNeglect手段にて無視するステップと、相互相関演算手段とNeglect手段の出力値から伝送路のパラメータをパラメータ推定手段にて推定するステップとを有し、互相関演算手段と自己相関演算手段を並列処理することにより高速演算を行うことを特徴とする信号分離方法であり、演算に時間のかかる自己相関演算と相互相関演算を並列処理することにより高速演算性、リアルタイム演算性を実現できる作用がある。
【００２１】
本発明の請求項７に記載の発明は、送信側から受信側へ信号を伝達する場合、伝達すべき所望信号（原信号）に対して、伝送路内で所望信号の識別に障害を来す雑音信号が任意の増幅率で重畳され、受信側において所望信号を識別できない場合において、受信側で雑音信号にうもれた所望信号を再生する為に、原信号発生手段から出力された原信号に対して第１及び第２のフィルタ手段にて任意タップ長原信号と雑音信号が重畳した信号から原信号のみを抽出する信号分離方法を記録した記憶媒体であって、原信号発生手段から出力された原信号に対して第１及び第２のフィルタ手段にて任意タップ長かつ任意タップ係数のフィルタリング処理を行うステップと、雑音発生手段から出力された雑音信号に対して第３及び第４のフィルタ手段にて任意タップ長かつ任意タップ係数のフィルタリング処理を行うステップと、第１のフィルタ手段の出力と第３のフィルタ手段の出力を第１の加算手段にて加算し、フィルタリングされた原信号と雑音を重ねあわせるステップと、第２のフィルタ手段と第４のフィルタ手段の出力を第２の加算手段にて加算し、フィルタリングされた雑音と原信号を重ねあわせるステップと、第１の加算手段及び第２の加算手段からの出力信号に対しフーリエ変換手段にて短時間フーリエ変換処理もしくはフーリエ変換処理を行うステップと、フーリエ変換手段からの出力値に対し相互相関演算手段にて相互相関演算を行うステップと、フーリエ変換手段からの出力値に対し自己相関演算手段にて自己相関演算を行うステップと、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値についてはNeglect手段にて無視するステップと、相互相関演算手段とNeglect手段の出力値から第１乃至第４のフィルタ手段及び第１、第２の加算手段を含む伝送路のパラメータをパラメータ推定手段にて推定するステップと、パラメータ推定手段で推定されたパラメータを元に伝送路の第１の加算手段からの出力値に対して第５のフィルタ手段にてフィルタリング処理を行うステップと、同じくパラメータ推定手段で推定されたパラメータを元に伝送路の第２の加算手段からの出力値に対して第６のフィルタ手段にてフィルタリング処理を行うステップと、伝送路の第１の加算手段からの出力値と第６のフィルタ手段の出力値を第３の加算手段にて加算するステップと、伝送路の第２の加算手段からの出力値と第５のフィルタ手段の出力値を第３の加算手段にて加算するステップと、第３の加算手段からの出力値に対して伝送路のパラメータを元に第１の乗算手段にて乗算処理を行うステップと、第４の加算手段からの出力値に対して伝送路のパラメータを元に第２の乗算手段にて乗算処理を行うステップとを実行させるためのプログラムを記録したことを特徴とする記録媒体であり、この記録したプログラムを実行することにより、理想的な信号分離装置の演算に最も近似した演算が行われるという作用がある。
【００２２】
本発明の請求項８に記載の発明は、２入力信号に対してフーリエ変換手段にて短時間フーリエ変換処理もしくはフーリエ変換処理を行うステップと、フーリエ変換手段からの出力値に対して相互相関演算手段にて相互相関演算を行うステップと、フーリエ変換手段からの出力値に対して自己相関演算手段にて自己相関演算を行うステップと、自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値についてはNeglect手段にて無視するステップと、相互相関演算手段とNeglect手段の出力値から伝送路のパラメータをパラメータ推定手段にて推定するステップとを実行させるためのプログラムを記録したことを特徴とする記録媒体であり、この記録したプログラムを実行することにより、理想的な信号分離装置の演算に最も近似した演算が行われるという作用がある。
【００２３】
以下、本発明の実施の形態について図１から図８を用いて説明する。
（実施の形態１）
図１は本発明の第１実施の形態による信号分離方式の機能ブロック図である。
【００２４】
図１において、送信側から受信側へ信号を伝達する場合、伝達すべき所望信号（原信号）に対して、伝送路内で所望信号の識別に障害を来す雑音信号が任意の増幅率で重畳され、受信側において所望信号を識別できない場合において、受信側で雑音信号にうもれた所望信号を再生する為に、１０は原信号を発生させる原信号発生手段、１１は原信号とは別に不必要な信号を発生させる雑音発生手段、１２、１４は原信号発生手段から出力された原信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行うフィルタ手段H11、H21、１３、１５は雑音発生手段から出力された雑音信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行うフィルタ手段H12、H22、１６はフィルタ手段１２の出力とフィルタ手段１３の出力を加算し、フィルタリングされた原信号と雑音を重ねあわせる加算手段、１７はフィルタ手段１４とフィルタ手段１５の出力を加算し、フィルタリングされた雑音と原信号を重ねあわせる加算手段、１８は加算手段１６、１７からの出力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段、１９はフーリエ変換手段１８からの出力値に対して相互相関演算を行う相互相関演算手段、２０はフーリエ変換手段１８からの出力値に対して自己相関演算を行う自己相関演算手段、２１は自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段、２２は相互相関演算手段１９とNeglect手段２１の出力値からフィルタ手段１２、フィルタ手段１３、フィルタ手段１４、フィルタ手段１５、及び加算手段１６，１７を含む伝送路のパラメータを推定するパラメータ推定手段、２３はパラメータ推定手段２２で推定されたパラメータを元に伝送路の加算手段１６からの出力値に対してフィルタリング処理を行うフィルタ手段G1、２４は同じくパラメータ推定手段２２で推定されたパラメータを元に伝送路の加算手段１７からの出力値に対してフィルタリング処理を行うフィルタ手段G2、２５は伝送路の加算手段１６からの出力値とフィルタ手段２４の出力値を加算する加算手段、２６は伝送路の加算手段１７からの出力値とフィルタ手段２３の出力値を加算する加算手段、２７は加算手段２５からの出力値に対して伝送路のパラメータを元に乗算処理を行う乗算手段、２８は加算手段２６からの出力値に対して伝送路のパラメータを元に乗算処理を行う乗算手段である。
【００２５】
次に図２を用いて、図１の実際の動作を説明する。
原信号発生手段１０で原信号（所望信号）が生成され、また雑音発生手段１１で雑音が発生する。原信号はフィルタ手段１２、フィルタ手段１４を通過し加算手段１６、１７で雑音成分を加算される。また雑音信号はフィルタ手段１３とフィルタ手段１５を通過し原信号成分と加算手段１６、１７で加算される（ステップ９）。
【００２６】
ここで、簡略化の為、原信号発生手段１０の出力値をZ変換しS₁(z)、雑音発生手段１１の出力値をZ変換しS₂(z)、加算手段１６の出力値をZ変換しY₁(z)、加算手段１７の出力値をZ変換しY₂(z)、加算手段２５の出力値をZ変換しV₁(z)、加算手段２６の出力値をZ変換しV₂(z)、フィルタ手段１２のインパルス応答をZ変換しH₁₁(z)、フィルタ手段１３のインパルス応答をZ変換しH₁₂(z)、フィルタ手段１４のインパルス応答をZ変換しH₂₁(z)、フィルタ手段１５のインパルス応答をZ変換しH₂₂(z)、フィルタ手段２３のインパルス応答をZ変換しG₁(z)、フィルタ手段２４のインパルス応答をZ変換しG₂(z)と表記する。
【００２７】
また、以上の表記を用いて各信号をベクトル表記すると（数２）、（数３）、（数４）が表現できる。
【００２８】
【数２】

【００２９】
【数３】

【００３０】
【数４】

【００３１】
また、（数２）、（数３）、（数４）からY(z)は（数５）で表現できる。
【００３２】
【数５】

【００３３】
フーリエ変換手段１８は、Y₁(z)、Y₂(z)の各々について短時間フーリエ変換処理もしくはフーリエ変換処理を行いY1(ω)、Y₂(ω)を算出する（ステップ１０）。
【００３４】
相互相関演算手段１９はフーリエ変換手段１８からの出力Y1(ω)、Y₂(ω)から相互相関演算を行う（ステップ１１）。
【００３５】
相互相関演算出力をP_y1y2、P_y2y1とすると、相互相関演算は（数６）、（数７）で表せられる。
【００３６】
【数６】

【００３７】
【数７】

【００３８】
自己相関演算手段２０はフーリエ変換手段１８からの出力Y1(ω)、Y₂(ω)から自己相関演算を行う（ステップ１２）。
【００３９】
自己相関演算出力をP_y1y1、P_y2y2とすると、自己相関演算は（数８）、（数９）で表せられる。
【００４０】
【数８】

【００４１】
【数９】

【００４２】
Neglect手段２１は自己相関演算出力P_y1y1、P_y2y2に対してある一定基準値以下の値を無視する（ステップ１３）。これはパラメータ推定手段２２での演算を安定させて行う為に必要である。
【００４３】
パラメータ推定手段２２は相互相関演算出力P_y1y2、P_y2y1とNeglectされた自己相関演算出力P_y1y1、P_y2y2からG₁(ω)、G₂(ω)を算出し、G₁(ω)、G₂(ω)を逆フーリエ変換処理してG₁(z)、G₂(z)を算出し、各々フィルタ手段２３とフィルタ手段２４のタップ係数値とする。
【００４４】
（ステップ１４）G₁(ω)、G₂(ω)の算出式は（数１０）、（数１１）で表現される。
【００４５】
【数１０】

【００４６】
【数１１】

【００４７】
フィルタ手段２３は加算手段１６からの出力値Y₁(z)に対して、タップ係数G₁(z)をもってフィルタリング処理を行い、また、フィルタ手段２４は加算手段１７からの出力値Y₂(z)に対して、タップ係数G₂(z)をもってフィルタリング処理を行う（ステップ１５）。
【００４８】
加算手段２５は加算手段１６とフィルタ手段２４からの出力値を加算し、加算手段２６は加算手段１７とフィルタ手段２３の出力値を加算する（ステップ１６）。
【００４９】
乗算手段２７は伝送路の各フィルタ手段１２，１３，１４，１５のパラメータを元に算出される1／Δを加算手段２５からの出力に対して乗算し、乗算手段２８は伝送路の各フィルタ手段１２，１３，１４，１５のパラメータを元に算出される1／Δを加算手段２６からの出力に対して乗算し（ステップ１７）、終了する（ステップ１８）。
【００５０】
1／Δは（数１２）により算出される。
【００５１】
【数１２】

【００５２】
本発明により、トレーニング信号を用いずに原信号と雑音を分離できる方式を提供できる。
【００５３】
（実施の形態２）
図３は本発明の第２実施の形態による信号分離方式の機能ブロック図である。
【００５４】
図３において、ツイストペアケーブルを用いて送信側から受信側へ信号を伝達する場合、ツイストペアケーブル内にて接地線に信号線の信号が任意の増幅率で重畳され送信側と受信側の接地電圧値に揺らぎが生じ、結果として受信側において所望信号を識別できない場合において、受信側で送信側と同じ接地電圧値を得る為に、４８は原信号を発生させる原信号発生手段、４９はデジタル信号をアナログ信号に変換するデジタル／アナログ変換手段、５０は原信号とは別に不必要な信号を発生させる雑音発生手段、５１、５３は原信号発生手段から出力された原信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行うフィルタ手段H11、H21、５２、５４は雑音発生手段から出力された雑音信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行うフィルタ手段H12、H22、５５はフィルタ手段５１の出力とフィルタ手段５２の出力を加算し、フィルタリングされた原信号と雑音を重ねあわせる加算手段、５６はフィルタ手段５３とフィルタ手段５４の出力を加算し、フィルタリングされた雑音と原信号を重ねあわせる加算手段、５７は加算手段５５からの出力アナログ信号をデジタル信号へ変換するアナログ／デジタル変換手段、５８は加算手段５６からの出力アナログ信号をデジタル信号へ変換するアナログ／デジタル変換手段、５９はアナログ／デジタル変換手段５７、５８からの出力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段、６０はフーリエ変換手段５９からの出力値に対して相互相関演算を行う相互相関演算手段、６１はフーリエ変換手段５９からの出力値に対して自己相関演算を行う自己相関演算手段、６２は自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段、６３は相互相関演算手段６０とNeglect手段６２の出力値からフィルタ手段５１、フィルタ手段５２、フィルタ手段５３、フィルタ手段５４、及び加算手段５５，５６を含むツイストペアケーブルのパラメータを推定するパラメータ推定手段、６４はパラメータ推定手段６３で推定されたパラメータを元にアナログ／デジタル変換手段５７からの出力値に対してフィルタリング処理を行うフィルタ手段G1、６５は同じくパラメータ推定手段６３で推定されたパラメータを元にアナログ／デジタル変換手段５８からの出力値に対してフィルタリング処理を行うフィルタ手段G2、６６はアナログ／デジタル変換手段５７からの出力値とフィルタ手段６５の出力値を加算する加算手段、６７はアナログ／デジタル変換手段５８からの出力値とフィルタ手段６４の出力値を加算する加算手段、６８は加算手段６６からの出力値に対してツイストペアケーブルのパラメータを元に乗算処理を行う乗算手段、６９は加算手段６７からの出力値に対してツイストペアケーブルのパラメータを元に乗算処理を行う乗算手段である。
【００５５】
次に図４を用いて、図３の実際の動作を説明する。
原信号発生手段４８で原信号（所望信号）が生成され、デジタル／アナログ変換手段４９でアナログ信号に変換される。また雑音発生手段５０で雑音が発生する。アナログ原信号はフィルタ手段５１、フィルタ手段５３を通過し加算手段５５、５６で雑音成分を加算される。また雑音信号はフィルタ手段５２とフィルタ手段５４を通過し原信号成分と加算手段５５、５６で加算される（ステップ２９）。加算手段５５、５６からのアナログ出力信号に対して各々アナログ／デジタル変換手段５７、５８がデジタル値に変換する（ステップ３０）。
【００５６】
ここで、簡略化の為、原信号発生手段４８の出力値をZ変換しS₁(z)、雑音発生手段５０の出力値をZ変換しS₂(z)、アナログ／デジタル変換手段５７の出力値をZ変換しY₁(z)、アナログ／デジタル変換手段５８の出力値をZ変換しY₂(z)、加算手段６６の出力値をZ変換しV₁(z)、加算手段６７の出力値をZ変換しV₂(z)、フィルタ手段５１のインパルス応答をZ変換しH₁₁(z)、フィルタ手段５２のインパルス応答をZ変換しH₁₂(z)、フィルタ手段５３のインパルス応答をZ変換しH₂₁(z)、フィルタ手段５４のインパルス応答をZ変換しH₂₂(z)、フィルタ手段６４のインパルス応答をZ変換しG₁(z)、フィルタ手段６５のインパルス応答をZ変換しG₂(z)と表記する。
【００５７】
また、以上の表記を用いて各信号をベクトル表記すると（数２）、（数３）、（数４）が表現できる。
【００５８】
また、（数２）、（数３）、（数４）からY(z)は（数５）で表現できる。
フーリエ変換手段５９は、Y₁(z)、Y₂(z)の各々について短時間フーリエ変換処理もしくはフーリエ変換処理を行いY1(ω)、Y₂(ω)を算出する（ステップ３１）。
【００５９】
相互相関演算手段６０はフーリエ変換手段５９からの出力Y1(ω)、Y₂(ω)から相互相関演算を行う（ステップ３２）。
【００６０】
相互相関演算出力をP_y1y2、P_y2y1とすると、相互相関演算は（数６）、（数７）で表せられる。
【００６１】
自己相関演算手段６１はフーリエ変換手段５９からの出力Y1(ω)、Y₂(ω)から自己相関演算を行う（ステップ３３）。
【００６２】
自己相関演算出力をP_y1y1、P_y2y2とすると、自己相関演算は（数８）、（数９）で表せられる。
【００６３】
Neglect手段６２は自己相関演算出力P_y1y1、P_y2y2に対してある一定基準値以下の値を無視する（ステップ３４）。これはパラメータ推定手段６３での演算を安定させて行う為に必要である。
【００６４】
パラメータ推定手段６３は相互相関演算出力P_y1y2、P_y2y1とNeglectされた自己相関演算出力P_y1y1、P_y2y2からG₁(ω)、G₂(ω)を算出し、G₁(ω)、G₂(ω)を逆フーリエ変換処理してG₁(z)、G₂(z)を算出し、各々フィルタ手段６４とフィルタ手段６５のタップ係数値とする（ステップ３５）。G₁(ω)、G₂(ω)の算出式は（数１０）、（数１１）で表現される。
【００６５】
フィルタ手段６４はアナログ／デジタル変換手段５７からの出力値Y₁(z)に対して、タップ係数G₁(z)をもってフィルタリング処理を行い、また、フィルタ手段６５はアナログ／デジタル変換手段５８からの出力値Y₂(z)に対して、タップ係数G₂(z)をもってフィルタリング処理を行う（ステップ３６）。
【００６６】
加算手段６６はアナログ／デジタル変換手段５７とフィルタ手段６５からの出力値を加算し、加算手段６７はアナログ／デジタル変換手段５８とフィルタ手段６４の出力値を加算する（ステップ３７）。
【００６７】
乗算手段６８はツイストペアケーブルの各フィルタ手段５１，５２，５３，５４のパラメータを元に算出される1／Δを加算手段６６からの出力に対して乗算し、乗算手段６９はツイストペアケーブルの各フィルタ手段５１，５２，５３，５４のパラメータを元に算出される1／Δを加算手段６７からの出力に対して乗算し（ステップ３８）、終了する（ステップ３９）。
【００６８】
1／Δは（数１２）により算出される。
本発明により、トレーニング信号を用いずにツイストペアケーブルによる通信において原信号と雑音を分離できる信号分離装置を実現できる。
【００６９】
（実施の形態３）
図５は本発明の第３実施の形態による信号分離方式の機能ブロック図である。
【００７０】
図５において、カーナビゲーションシステムの音声認識装置や携帯型電話機、ＯＦＤＭ（Orthogonal Frequency Division Multiplexing : 直交周波数分割多重方式）通信等において、カーナビゲーションシステムの音声認識装置では話者（送信側）からカーナビゲーション装置（受信側）へ音声コマンドを伝達する場合、話者の発した音声コマンド（原信号）に対して、車内のエンジン装置やタイヤ装置の路面との摩擦音等の音声コマンドの識別に障害を来す雑音信号が任意の増幅率で重畳され、カーナビゲーション装置において音声コマンドを識別できない場合において、カーナビゲーション装置で雑音に埋もれた音声コマンドを再生する為に、また、携帯電話装置では話者の音声信号（原信号）に対して話者以外の周囲雑音（雑音信号）が任意の増幅率で重畳され、話者の通話者では音声を認識できない場合において、送信側の携帯電話装置において雑音に埋もれた話者の声（原信号）を再生する為に、また、ＯＦＤＭ通信においては複数の送信局から発信された信号（原信号を受信局にて所望局の信号（原信号）のみを取り出す為に本発明を応用したものである。図において、７０は原信号を発生させる原信号発生手段、７１は原信号とは別に不必要な信号を発生させる雑音発生手段、７２、７４は原信号発生手段から出力された原信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行うフィルタ手段H11、H21、７３、７５は雑音発生手段から出力された雑音信号に対して任意タップ長かつ任意タップ係数のフィルタリング処理を行うフィルタ手段H12、H22、７６はフィルタ手段７２の出力とフィルタ手段７３の出力を加算し、フィルタリングされた原信号と雑音を重ねあわせる加算手段、７７はフィルタ手段７４とフィルタ手段７５の出力を加算し、フィルタリングされた雑音と原信号を重ねあわせる加算手段、７８は加算手段７６からの信号を電気信号へ変換するマイクロフォン、７９は加算手段７７からの信号を電気信号へ変換するマイクロフォン、８０はマイクロフォン７８からの出力アナログ電気信号をデジタル信号へ変換するアナログ／デジタル変換手段、８１はマイクロフォン７９からの出力アナログ電気信号をデジタル信号へ変換するアナログ／デジタル変換手段、８２はアナログ／デジタル変換手段８０、８１からの出力信号に対して短時間フーリエ変換処理もしくはフーリエ変換処理を行うフーリエ変換手段、８３はフーリエ変換手段８２からの出力値に対して相互相関演算を行う相互相関演算手段、８４はフーリエ変換手段８２からの出力値に対して自己相関演算を行う自己相関演算手段、８５は自己相関演算手段からの出力値に対して振幅がある一定の基準値以上の値については無視するNeglect手段、８６は相互相関演算手段８３とNeglect手段８５の出力値からフィルタ手段７２、フィルタ手段７３、フィルタ手段７４、フィルタ手段７５、及び加算手段７６，７７を含む伝送路のパラメータを推定するパラメータ推定手段、８７はパラメータ推定手段８６で推定されたパラメータを元にアナログ／デジタル変換手段８０からの出力値に対してフィルタリング処理を行うフィルタ手段G1、８８は同じくパラメータ推定手段８６で推定されたパラメータを元にアナログ／デジタル変換手段８１からの出力値に対してフィルタリング処理を行うフィルタ手段G2、８９はアナログ／デジタル変換手段８０からの出力値とフィルタ手段８８の出力値を加算する加算手段、９０はアナログ／デジタル変換手段８１からの出力値とフィルタ手段８７の出力値を加算する加算手段、９１は加算手段８９からの出力値に対して伝送路のパラメータを元に乗算処理を行う乗算手段、９２は加算手段９０からの出力値に対して伝送路のパラメータを元に乗算処理を行う乗算手段である。
【００７１】
次に図６を用いて、図５の実際の動作を説明する。
原信号発生手段７０で原信号（所望信号）が生成され、また雑音発生手段７１で雑音が発生する。原信号はフィルタ手段７２、フィルタ手段７４を通過し加算手段７６、７７で雑音成分を加算される。また雑音信号はフィルタ手段７３とフィルタ手段７５を通過し原信号成分と加算手段７６、７７で加算される（ステップ４０）。加算手段７６、７７からの音声信号に対して各々マイクロフォン７８、７９が音声信号を電気信号に変換する（ステップ４１）。マイクロフォン７８、７９からのアナログ電気信号に対して各々アナログ／デジタル変換手段８０、８１がデジタル値に変換する（ステップ４２）。
【００７２】
ここで、簡略化の為、原信号発生手段７０の出力値をZ変換しS₁(z)、雑音発生手段７１の出力値をZ変換しS₂(z)、アナログ／デジタル変換手段８０の出力値をZ変換しY₁(z)、アナログ／デジタル変換手段８１の出力値をZ変換しY₂(z)、加算手段８９の出力値をZ変換しV₁(z)、加算手段９０の出力値をZ変換しV₂(z)、フィルタ手段７２のインパルス応答をZ変換しH₁₁(z)、フィルタ手段７３のインパルス応答をZ変換しH₁₂(z)、フィルタ手段７４のインパルス応答をZ変換しH₂₁(z)、フィルタ手段７５のインパルス応答をZ変換しH₂₂(z)、フィルタ手段８７のインパルス応答をZ変換しG₁(z)、フィルタ手段８８のインパルス応答をZ変換しG₂(z)と表記する。
【００７３】
また、以上の表記を用いて各信号をベクトル表記すると（数２）、（数３）、（数４）が表現できる。
【００７４】
また、（数２）、（数３）、（数４）からY(z)は（数５）で表現できる。
フーリエ変換手段８２は、Y₁(z)、Y₂(z)の各々について短時間フーリエ変換処理もしくはフーリエ変換処理を行いY1(ω)、Y₂(ω)を算出する（ステップ４３）。
【００７５】
相互相関演算手段８３はフーリエ変換手段８２からの出力Y1(ω)、Y₂(ω)から相互相関演算を行う（ステップ４４）。
【００７６】
相互相関演算出力をP_y1y2、P_y2y1とすると、相互相関演算は（数６）、（数７）で表せられる。
【００７７】
自己相関演算手段８４はフーリエ変換手段８２からの出力Y1(ω)、Y₂(ω)から自己相関演算を行う（ステップ４５）。
【００７８】
自己相関演算出力をP_y1y1、P_y2y2とすると、自己相関演算は（数８）、（数９）で表せられる。
【００７９】
Neglect手段８５は自己相関演算出力P_y1y1、P_y2y2に対してある一定基準値以下の値を無視する（ステップ４６）。これはパラメータ推定手段８６での演算を安定させて行う為に必要である。
【００８０】
パラメータ推定手段８６は相互相関演算出力P_y1y2、P_y2y1とNeglectされた自己相関演算出力P_y1y1、P_y2y2からG₁(ω)、G₂(ω)を算出し、G₁(ω)、G₂(ω)を逆フーリエ変換処理してG₁(z)、G₂(z)を算出し、各々フィルタ手段８７とフィルタ手段８８のタップ係数値とする。（ステップ４７）G₁(ω)、G₂(ω)の算出式は（数１０）、（数１１）で表現される。
【００８１】
フィルタ手段８７はアナログ／デジタル変換手段８０からの出力値Y₁(z)に対して、タップ係数G₁(z)をもってフィルタリング処理を行い、また、フィルタ手段８８はアナログ／デジタル変換手段８１からの出力値Y₂(z)に対して、タップ係数G₂(z)をもってフィルタリング処理を行う（ステップ４８）。
【００８２】
加算手段８９はアナログ／デジタル変換手段８０とフィルタ手段８８からの出力値を加算し、加算手段９０はアナログ／デジタル変換手段８１とフィルタ手段８７の出力値を加算する（ステップ４９）。
【００８３】
乗算手段９１は伝送路の各フィルタ手段７２，７３，７４，７５のパラメータを元に算出される1／Δを加算手段８９からの出力に対して乗算し、乗算手段９２は伝送路の各フィルタ手段７２，７３，７４，７５のパラメータを元に算出される1／Δを加算手段９０からの出力に対して乗算し（ステップ５０）、終了する（ステップ５１）。
【００８４】
1／Δは（数１２）により算出される。
本発明により、トレーニング信号を用いずにカーナビゲーションシステムの音声認識手段や携帯型電話機、ＯＦＤＭ通信等において周囲の雑音（ＯＦＤＭ通信においては他チャンネルの信号）を取り除き所望信号のみ抽出できる信号分離装置を実現できる。
【００８５】
（実施の形態４）
図７は本発明の第４実施の形態による信号分離方式の機能ブロック図である。
【００８６】
９３は入力信号に対して短時間フーリエ変換処理、もしくはフーリエ変換処理を行うフーリエ変換手段、９４はフーリエ変換手段９３からの出力信号から相互相関演算を行う相互相関演算手段、９５はフーリエ変換手段９３からの出力信号から自己相関演算を行う自己相関演算手段、９６は自己相関演算手段９５からの出力信号に対してある一定基準値以下の値は無視するNeglect手段、９７は相互相関演算手段９４の出力信号とNeglect手段９６からの出力信号から伝送路のパラメータを推定するパラメータ推定手段である。
【００８７】
次に図８を用いて、図７の実際の動作を説明する。
フーリエ変換手段９３に信号が入力されると（ステップ５２）、フーリエ変換手段９３が２系統の入力信号に対して各々に短時間フーリエ変換演算、もしくは、フーリエ変換演算を行う（ステップ５３）。フーリエ変換手段からの出力信号に対して相互相関演算手段が相互相関演算を行い（ステップ５４）、同じく自己相関演算手段が自己相関演算を行う（ステップ５５）。ステップ５４とステップ５５に関しては演算に時間がかかる為、ハードウェアとして独立した手段を組み、ステップ５４とステップ５５は並列に演算を行う。
【００８８】
Neglect手段９６は自己相関演算手段９５からの出力信号に対して、ある一定基準値以下の値は無視し（ステップ５６）、パラメータ推定手段は相互相関演算手段９４からの出力信号とNeglect手段９６からの出力信号から（数１０）、（数１１）に示される演算を行い、その結果の逆フーリエ変換を行い（ステップ５７）、終了する（ステップ５８）。
【００８９】
【発明の効果】
以上のように本発明によれば、トレーニング信号を用いずに原信号（所望信号）と雑音を分離できるという作用を有しているため、通信においてはスループットが向上し、また、伝送路が時間的に変化する場合についても変化に追従して原信号（所望信号）と雑音を分離できる。また、カーナビゲーション装置においては車内雑音を取り除いた音声コマンドを受信できる為、音声認識率が向上し、携帯電話装置においては話者の周囲雑音を取り除ける為、雑音の無い高品質な通話が可能となり、ＯＦＤＭ通信では所望局のみの信号が高品質に受信可能となる効果を有する。
【図面の簡単な説明】
【図１】本発明の第１実施の形態による信号分離方式の機能ブロック図
【図２】本発明の第１実施の形態による信号分離方式の動作を示すフローチャート
【図３】本発明の第２実施の形態による信号分離装置の機能ブロック図
【図４】本発明の第２実施の形態による信号分離装置の動作を示すフローチャート
【図５】本発明の第３実施の形態による信号分離装置の機能ブロック図
【図６】本発明の第３実施の形態による信号分離装置の動作を示すフローチャート
【図７】本発明の第４実施の形態による信号分離装置の機能ブロック図
【図８】本発明の第４実施の形態による信号分離装置の動作を示すフローチャート
【図９】従来の技術による信号分離装置の機能ブロック図
【図１０】従来の技術による信号分離装置の動作を示すフローチャート
【符号の説明】
１ 原信号発生手段
２ トレーニング信号記憶手段
３ 切り替え手段
４ 雑音発生手段
５ 加算手段
６ 伝送路フィルタ手段
７ フィルタ処理手段
８ トレーニング信号記憶手段
９ 伝送路逆特性演算手段
１０ 原信号発生手段
１１ 雑音発生手段
１２ フィルタ手段
１３ フィルタ手段
１４ フィルタ手段
１５ フィルタ手段
１６ 加算手段
１７ 加算手段
１８ フーリエ変換手段
１９ 相互相関演算手段
２０ 自己相関演算手段
２１ Neglect手段
２２ パラメータ推定手段
２３ フィルタ手段
２４ フィルタ手段
２５ 加算手段
２６ 加算手段
２７ 乗算手段
２８ 乗算手段
４８ 原信号発生手段
４９ デジタル／アナログ変換手段
５０ 雑音発生手段
５１ フィルタ手段
５２ フィルタ手段
５３ フィルタ手段
５４ フィルタ手段
５５ 加算手段
５６ 加算手段
５７ アナログ／デジタル変換手段
５８ アナログ／デジタル変換手段
５９ フーリエ変換手段
６０ 相互相関演算手段
６１ 自己相関演算手段
６２ Neglect手段
６３ パラメータ推定手段
６４ フィルタ手段
６５ フィルタ手段
６６ 加算手段
６７ 加算手段
６８ 乗算手段
６９ 乗算手段
７０ 原信号発生手段
７１ 雑音発生手段
７２ フィルタ手段
７３ フィルタ手段
７４ フィルタ手段
７５ フィルタ手段
７６ 加算手段
７７ 加算手段
７８ マイクロフォン
７９ マイクロフォン
８０ アナログ／デジタル変換手段
８１ アナログ／デジタル変換手段
８２ フーリエ変換手段
８３ 相互相関演算手段
８４ 自己相関演算手段
８５ Neglect手段
８６ パラメータ推定手段
８７ フィルタ手段
８８ フィルタ手段
８９ 加算手段
９０ 加算手段
９１ 乗算手段
９２ 乗算手段
９３ フーリエ変換手段
９４ 相互相関演算手段
９５ 自己相関演算手段
９６ Neglect手段
９７ パラメータ推定手段[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a signal separation apparatus, a signal separation method, and a recording medium that perform signal processing, and in particular, perform signal separation processing.
[0002]
[Prior art]
There is a signal separation processing technique that removes noise from a transmitted signal and restores the original signal by processing on the receiving side. According to this signal separation processing technique, a signal in a state where noise is added in the transmission path and the original signal cannot be discriminated can be brought close to the original original signal. A conventional signal separation processing technique will be described below. FIG. 9 is a functional block diagram of a conventional signal separation apparatus.
[0003]
In FIG. 9, 1 is an original signal generating means for generating an original signal, 2 is a training signal storing means for storing a training signal, 3 is a signal from the original signal generating means 1 and a signal from the training signal storing means 2 Switching means for switching signals, 4 is noise generating means generated in the transmission line, 5 is an adding means for adding the output signal from the switching means 3 and the output signal from the noise generating means 4, and 6 is for attenuating the signal in the transmission line A transmission path filter means for giving a distortion, a filter for outputting a signal close to the original signal from which noise has been removed by filtering the output signal from the transmission path including the noise generation means, the addition means, and the transmission path filter means A processing means, 8 is a training signal storage means for storing a training signal, 9 is an output signal from the transmission path and a filter processing means 7 It calculates the inverse characteristics of the transmission path from the output signal from the al and the output signal from the training signal memory means 8, a transmission path inverse characteristics calculating means for calculating the tap coefficients of the filtering means 7.
[0004]
The operation of the signal separation device in the prior art of FIG. 9 will be described with reference to FIG.
[0005]
The switching means 3 outputs the signal from the training signal storage means 2 to the adding means 5, and the noise from the noise generating means 4 and the output signal from the switching means 3 are added to give attenuation / distortion by the transmission line filter means ( Step 1). A signal that is overlapped by noise and further attenuated / distorted by the transmission path filter means is input to the filter processing means 7 and the transmission path inverse characteristic calculation means 8 (step 2).
[0006]
Here, the output signal from the transmission line filter means 6 is X (n) at time n, the output signal from the filter processing means 7 is Y (n), the output signal from the training signal storage is D (n), and the reverse of the transmission line. The tap coefficient value of the reverse characteristic of the transmission path calculated by the characteristic calculation means 8 is expressed as C (n).
[0007]
The transmission line reverse characteristic calculation means 9 calculates the reverse characteristic of the transmission line from X (n), Y (n), and D (n) (step 3). Various methods for calculating the inverse characteristic have been proposed, but a typical example is an LMS (Least Mean Square) method.
[0008]
In the LMS system, the filter tap coefficient value C (n) of the reverse characteristic of the transmission path is expressed by (Equation 1).
[0009]
[Expression 1]

[0010]
The filter processing means 7 performs a filtering process on the output signal X (n) from the transmission path filter means 6 based on the tap coefficient C (n) calculated by the transmission path inverse characteristic calculation means 9 (step 4).
[0011]
The transmission path inverse characteristic value C (n) is learned by performing the above steps a specified number of times, and when the reception of the training signal is completed a specified number of times (step 5), the switching means 3 receives a signal from the original signal generating means. Is output to the adding means 5 (step 6). The signal output to the adding means 5 in step 6 is added to the noise signal from the noise generating means 4 by the adding means 5 and further attenuated / distorted by the transmission line filter means 6. The filter processing means 7 performs a filtering process on the output signal from the transmission path filter means 6 using the transmission path inverse characteristic C (n) learned at the time of receiving the training signal (step 7), and the output from the transmission path filter means A signal Y (n) close to the original signal obtained by removing noise from the signal is output, and the process ends (step 8).
[0012]
In Step 6, Step 7, and Step 8, the transmission path inverse characteristic value C (n) is not calculated.
[0013]
[Problems to be solved by the invention]
However, in the above configuration, it is necessary to transmit and receive a training signal in addition to the communication using the original signal, and the throughput is reduced. Further, since the transmission path inverse characteristic value C (n) is not calculated in Step 6, Step 7 and Step 8, there is a problem that it is not possible to cope with the case where the characteristics of the transmission path change with time during transmission and reception of the original signal. Was.
[0014]
[Means for Solving the Problems]
In order to solve the above-described problems, the present invention provides a noise signal that hinders identification of a desired signal in a transmission path with respect to a desired signal (original signal) to be transmitted when the signal is transmitted from the transmitting side to the receiving side. Is superimposed at an arbitrary amplification factor and the desired signal cannot be identified on the receiving side, the short-time Fourier transform processing or Fourier transform is performed on the two input signals in order to reproduce the desired signal covered with the noise signal on the receiving side. Fourier transform means for performing processing, cross-correlation calculation means for performing cross-correlation calculation on the output value from the Fourier transform means, auto-correlation calculation means for performing auto-correlation calculation on the output value from the Fourier transform means, The neglect means that ignores values greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculation means, and the parameters of the transmission path from the output values of the cross correlation calculation means and the neglect means. Parameter estimation means to estimate the data, so that the original signal and noise can be separated without the need for a training signal, and the transmission path changes even when the transmission path changes over time during transmission and reception of the original signal An algorithm capable of performing signal separation processing following the above can be provided.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
According to the first aspect of the present invention, when a signal is transmitted from the transmitting side to the receiving side, the desired signal to be transmitted (original signal) is disturbed in identifying the desired signal in the transmission path. When the noise signal is superimposed at an arbitrary amplification factor and the desired signal cannot be identified on the receiving side, the original signal output from the original signal generating means is used to reproduce the desired signal covered by the noise signal on the receiving side. First and second filter means for filtering arbitrary tap length and arbitrary tap coefficient, and third and second filtering means for filtering arbitrary tap length and arbitrary tap coefficient for the noise signal output from the noise generating means. A fourth filter means, a first addition means for adding the outputs of the first filter means and the output of the third filter means, and superimposing the filtered original signal and noise; and a second filter. A second adding means for adding the outputs of the filter means and the fourth filter means and superimposing the filtered noise and the original signal; and a short-time Fourier transform process or Fourier transform for the output signal from the first adding means. Fourier transform means for performing transformation processing, cross-correlation calculation means for performing cross-correlation calculation on the output value from the Fourier transform means, and auto-correlation calculation means for performing autocorrelation calculation on the output value from the Fourier transform means; , Neglect means for ignoring a value greater than a certain reference value with respect to the output value from the autocorrelation calculation means, first to fourth filter means from the output values of the cross correlation calculation means and the neglect means, and Parameter estimating means for estimating the parameters of the transmission line including the first and second adding means, and the first of the transmission lines based on the parameters estimated by the parameter estimating means A fifth filtering means that performs filtering on the output value from the adding means, and filtering processing on the output value from the second adding means on the transmission line based on the parameter estimated by the parameter estimating means. 6th filter means to perform, 3rd addition means which adds the output value from the 1st addition means of a transmission line, and the output value of the 6th filter means, and the output from the 2nd addition means of a transmission line A fourth addition means for adding the value and the output value of the fifth filter means; a first multiplication means for performing a multiplication process on the output value from the third addition means based on the parameters of the transmission line; A signal separation device having second multiplication means for performing a multiplication process on the output value from the fourth addition means based on the transmission line parameter, and to obtain the output value from the transmission line without requiring a training signal. In contrast to the original signal and noise Performs signal separation processing of separating an effect of extracting a signal close to the original signal.
[0016]
According to the second aspect of the present invention, when a signal is transmitted from the transmitting side to the receiving side using a twisted pair cable, the signal of the signal line is superimposed on the ground line within the twisted pair cable at an arbitrary amplification factor and transmitted. In order to obtain the same ground voltage value as that of the transmitting side on the receiving side when the desired signal cannot be identified on the receiving side as a result of fluctuations in the ground voltage value on the receiving side and the receiving side, Digital / analog conversion means for converting a digital output signal from the original signal generation means as a voltage value into an analog signal, and filtering processing of an arbitrary tap length and an arbitrary tap coefficient for the signal output from the digital / analog conversion means For the noise signal output from the first and second filter means to be performed and the noise generating means as the transmission side signal line The third and fourth filter means for filtering the arbitrary tap length and arbitrary tap coefficients, the output of the first filter means and the output of the third filter means are added, and the filtered analog original signal and noise are added. From the first addition means for superimposing, the second addition means for adding the outputs of the second filter means and the fourth filter means, and superposing the filtered noise and the analog original signal, and the first addition means First analog / digital conversion means for converting the analog output signal into a digital signal, second analog / digital conversion means for converting the analog output signal from the second addition means into a digital signal, first and first Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on the output signal from the analog / digital conversion means 2 , A cross-correlation calculating means for performing a cross-correlation calculation on the output value from the Fourier transform means, an auto-correlation calculating means for performing an auto-correlation calculation on the output value from the Fourier transform means, and an output from the auto-correlation calculating means Neglect means for ignoring a value greater than a certain reference value with respect to the value, first to fourth filter means and first and second addition means from the output values of the cross-correlation calculating means and Neglect means Parameter estimating means for estimating the parameters of the transmission line including the fifth filter means for performing filtering on the output value from the first adding means of the twisted pair cable based on the parameter estimated by the parameter estimating means; Similarly, filtering processing is performed on the output value from the second addition means of the twisted pair cable based on the parameter estimated by the parameter estimation means. A sixth adding means for performing, a third adding means for adding the output value from the adding means of the twisted pair cable and the output value of the sixth filter means, an output value from the second adding means of the twisted pair cable, and the second A fourth adding means for adding the output values of the filter means, a first multiplying means for performing a multiplication process on the output value from the third adding means based on the parameters of the twisted pair cable, A signal separation device having a second multiplying unit that multiplies an output value from the adding unit based on a twisted pair cable parameter, and outputs from the twisted pair cable without requiring a training signal in communication using the twisted pair cable. A signal separation process that separates the original signal and noise from the value is performed to extract a signal close to the original signal, and the noise is detected from the ground voltage value on the receiving side Min was removed, it has an effect of realizing equally be stable signal transmission and the transmission side.
[0017]
The invention according to claim 3 of the present invention is a speech recognition device for a car navigation system, such as a speech recognition device for a car navigation system, a portable telephone, or OFDM (Orthogonal Frequency Division Multiplexing) communication. When a voice command is transmitted from a person (transmission side) to a car navigation device (reception side), the voice command (original signal) is identified with respect to a voice command such as a friction sound with the engine device in the vehicle or the road surface of the tire device. In order to reproduce a voice command buried in noise in a car navigation device when a noise signal that causes an obstacle is superimposed at an arbitrary amplification factor and the voice command cannot be identified in the car navigation device, it is also possible to talk on a mobile phone device. Ambient noise other than the speaker is superimposed on the voice signal of the speaker (original signal) with an arbitrary amplification factor In order to reproduce the voice of the speaker buried in noise in the mobile phone device on the transmission side when the speaker's caller cannot recognize the voice, the signal from a plurality of transmission stations is used in the OFDM communication. In order to extract only the signal (original signal) of the desired station at the receiving station, when extracting only the desired signal from the desired signal mixed with noise (the signal of the other channel in OFDM communication), it is output from the original signal generating means. First and second filter means for filtering arbitrary tap length and arbitrary tap coefficient for the received signal, and filtering processing of arbitrary tap length and arbitrary tap coefficient for the noise signal output from the noise generating means The third and fourth filter means to be performed, the output of the first filter means and the output of the third filter means are added, and the filtered original signal and noise are added. First addition means for superimposing sounds, second addition means for adding the outputs of the second filter means and the fourth filter means, and superposing the filtered noise and the original signal; and first addition means The first microphone for converting the output signal from the first microphone into an electric signal, the second microphone for converting the output signal from the second adding means into the electric signal, and the analog output signal from the first microphone as digital First analog / digital conversion means for converting into a signal, second analog / digital conversion means for converting an analog output signal from the second microphone into a digital signal, and first and second analog / digital conversion means Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on the output signal from, and the output value from the Fourier transform means A cross-correlation calculating means for performing a cross-correlation calculation, an auto-correlation calculating means for performing an auto-correlation operation on the output value from the Fourier transform means, and an output value from the auto-correlation calculating means having a certain amplitude. Neglect means for ignoring values above the reference value, and the parameters of the transmission line including the first to fourth filter means and the first and second addition means are estimated from the output values of the cross correlation calculation means and the neglect means. A parameter estimation unit, a fifth filter unit that performs filtering on the output value from the first addition unit of the transmission line based on the parameter estimated by the parameter estimation unit, and also estimated by the parameter estimation unit Sixth filter means for performing filtering processing on the output value from the second addition means of the transmission line based on the parameter, and output from the first addition means of the transmission line A third adding means for adding the value and the output value of the sixth filter means; a fourth adding means for adding the output value from the second adding means of the transmission line and the output value of the fifth filter means; The first multiplication means for performing the multiplication process on the output value from the third addition means based on the transmission path parameter, and the transmission value on the output value from the fourth addition means based on the transmission path parameter A signal separation device having second multiplication means for performing multiplication processing, and performing signal separation processing for separating an original signal and noise on an output value from a transmission line without requiring a training signal, and a signal close to the original signal Has the effect of extracting
[0018]
According to a fourth aspect of the present invention, a Fourier transform unit that performs a short-time Fourier transform process or a Fourier transform process on two input signals, and a mutual correlation operation that performs a cross-correlation operation on an output value from the Fourier transform unit. Correlation calculation means, autocorrelation calculation means for performing autocorrelation calculation on the output value from Fourier transform means, and values exceeding an amplitude with respect to the output value from autocorrelation calculation means are ignored. Neglect means, cross-correlation computing means, and parameter estimation means for estimating transmission path parameters from the output values of Neglect means, and performing high-speed computation by parallel processing of cross-correlation computing means and autocorrelation computing means This is a signal separation device characterized by high-speed and real-time operability by performing parallel processing of autocorrelation and cross-correlation operations that take time. There is an effect that can be achieved.
[0019]
According to the fifth aspect of the present invention, when a signal is transmitted from the transmitting side to the receiving side, the desired signal to be transmitted (original signal) is disturbed in identifying the desired signal in the transmission path. When the noise signal is superimposed at an arbitrary amplification factor and the desired signal cannot be identified on the receiving side, the original signal output from the original signal generating means is used to reproduce the desired signal covered by the noise signal on the receiving side. Filtering the arbitrary tap length and arbitrary tap coefficient by the first and second filter means, and the arbitrary tap by the third and fourth filter means for the noise signal output from the noise generating means The filtering process of a long and arbitrary tap coefficient, the output of the first filter means and the output of the third filter means are added by the first adding means, and the filtered original signal and noise are mixed. Superimposing, adding the outputs of the second filter means and the fourth filter means by the second adding means, superimposing the filtered noise and the original signal, the first adding means and the first adding means A step of performing a short-time Fourier transform process or a Fourier transform process on the output signal from the addition means of 2 in the Fourier transform means, and a step of performing a cross-correlation operation on the output value from the Fourier transform means in the cross-correlation operation means A step of performing autocorrelation calculation on the output value from the Fourier transform means by the autocorrelation calculation means, and Neglect means for a value greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculation means And the first to fourth filter means and the first and second adding means are transmitted from the output values of the cross-correlation calculating means and the neglect means. A step of estimating the parameter of the transmission path by the parameter estimation means, and a filtering process by the fifth filter means for the output value from the first addition means of the transmission path based on the parameter estimated by the parameter estimation means Performing a filtering process by the sixth filter means on the output value from the second addition means of the transmission line based on the parameter estimated by the parameter estimation means, A step of adding the output value from the first adding means and the output value of the sixth filter means by the third adding means; an output value from the second adding means of the transmission line; and an output of the fifth filter means A step of adding a value by a third addition unit, a step of performing a multiplication process by a first multiplication unit on the basis of a transmission path parameter for an output value from the third addition unit, And a step of multiplying the output value from the addition means by the second multiplication means on the basis of the transmission line parameters, and output from the transmission line without requiring a training signal. There is an effect of extracting a signal close to the original signal by performing signal separation processing for separating the original signal and noise on the value.
[0020]
According to a sixth aspect of the present invention, a step of performing a short-time Fourier transform process or a Fourier transform process on a two-input signal by a Fourier transform unit, and a cross-correlation operation on an output value from the Fourier transform unit A step of performing a cross-correlation calculation by the means, a step of performing an autocorrelation calculation by the autocorrelation calculation means for the output value from the Fourier transform means, and a constant amplitude for the output value from the autocorrelation calculation means And a step of ignoring the value above the reference value by the neglect means, and a step of estimating the parameter of the transmission path by the parameter estimation means from the output value of the cross correlation calculation means and the neglect means, and the cross correlation calculation means Is a signal separation method characterized in that high-speed computation is performed by parallel processing of the autocorrelation computation means and the autocorrelation computation means. Fast calculation of by parallel processing operations, an effect that can realize real-time calculation of.
[0021]
According to the seventh aspect of the present invention, when a signal is transmitted from the transmission side to the reception side, the desired signal to be transmitted (original signal) is disturbed in identifying the desired signal in the transmission path. When the noise signal is superimposed at an arbitrary amplification factor and the desired signal cannot be identified on the receiving side, the original signal output from the original signal generating means is used to reproduce the desired signal covered by the noise signal on the receiving side. A storage medium storing a signal separation method for extracting only the original signal from the signal in which the arbitrary tap length original signal and the noise signal are superimposed by the first and second filter means, and the original output from the original signal generating means Filtering the signal with an arbitrary tap length and an arbitrary tap coefficient by the first and second filter means, and applying the third and fourth filter means to the noise signal output from the noise generating means. A step of filtering arbitrary tap length and arbitrary tap coefficient, and the output of the first filter means and the output of the third filter means are added by the first adding means, and the filtered original signal and noise are superimposed. Combining the outputs of the second filter means and the fourth filter means by the second addition means, and superimposing the filtered noise and the original signal; the first addition means and the second addition means; Performing a short-time Fourier transform process or a Fourier transform process on the output signal from the adding means in the Fourier transform means; and performing a cross-correlation operation on the output value from the Fourier transform means in the cross-correlation calculating means; A step of performing autocorrelation calculation on the output value from the Fourier transform means by the autocorrelation calculation means, and an output from the autocorrelation calculation means For a value greater than a certain reference value, the neglect means ignores the first to fourth filter means and the first and second filter means from the output values of the cross-correlation calculating means and the neglect means. A step of estimating the parameter of the transmission line including the adding means by the parameter estimating means; and a fifth filter means for the output value from the first adding means of the transmission line based on the parameter estimated by the parameter estimating means Performing a filtering process at the sixth filtering means on the output value from the second adding means of the transmission line based on the parameter estimated by the parameter estimating means, A step of adding the output value from the first addition means of the transmission line and the output value of the sixth filter means by the third addition means, and a step from the second addition means of the transmission line. The step of adding the force value and the output value of the fifth filter means by the third addition means, and the output value from the third addition means by the first multiplication means based on the parameters of the transmission path Recording a program for executing the step of performing the multiplication process and the step of performing the multiplication process by the second multiplication unit based on the parameters of the transmission path with respect to the output value from the fourth addition unit; This is a characteristic recording medium, and by executing this recorded program, there is an effect that the operation most similar to the operation of the ideal signal separation device is performed.
[0022]
According to an eighth aspect of the present invention, a step of performing a short-time Fourier transform process or a Fourier transform process on a two-input signal by a Fourier transform unit, and a cross-correlation operation on an output value from the Fourier transform unit A step of performing a cross-correlation calculation by the means, a step of performing an autocorrelation calculation by the autocorrelation calculation means for the output value from the Fourier transform means, and a constant amplitude for the output value from the autocorrelation calculation means Records a program to execute a step of ignoring the value above the reference value by the neglect unit and a step of estimating the parameter of the transmission path by the parameter estimation unit from the output value of the cross-correlation calculating unit and the neglect unit By executing this recorded program, it is the closest to the operation of an ideal signal separation device. There is effect that operation is performed.
[0023]
Hereinafter, embodiments of the present invention will be described with reference to FIGS.
(Embodiment 1)
FIG. 1 is a functional block diagram of a signal separation system according to the first embodiment of the present invention.
[0024]
In FIG. 1, when a signal is transmitted from a transmission side to a reception side, a noise signal that impedes identification of the desired signal in the transmission path with respect to the desired signal (original signal) to be transmitted has an arbitrary amplification factor. In the case where the desired signal cannot be identified on the receiving side, 10 is an original signal generating means for generating the original signal, and 11 is a separate signal from the original signal. Noise generating means for generating necessary signals, 12 and 14 are filter means H11, H21, 13, and 15 for performing filtering processing of arbitrary tap lengths and arbitrary tap coefficients on the original signal output from the original signal generating means. Filter means H12, H22, and 16 for performing filtering processing with an arbitrary tap length and an arbitrary tap coefficient on the noise signal output from the generating means are the outputs of the filter means 12 and the filter means 13. Adding means for adding the force and superimposing the filtered original signal and the noise; 17 adding the outputs of the filter means 14 and the filter means 15; and adding means for superposing the filtered noise and the original signal; Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on the output signals from 16 and 17, 19 is a cross-correlation computing means for performing a cross-correlation operation on the output value from the Fourier transform means 18, and 20 An autocorrelation calculating means for performing autocorrelation calculation on the output value from the Fourier transform means 18, and a neglect means for ignoring a value greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculation means , 22 are filter means 12, filter means 13, and filter means 14 from the output values of the cross-correlation calculating means 19 and neglect means 21. , A parameter estimation means for estimating the parameters of the transmission line including the filter means 15 and the addition means 16 and 17, and 23 for the output value from the addition means 16 of the transmission path based on the parameters estimated by the parameter estimation means 22. Similarly, the filtering means G1 and 24 that perform filtering processing are filtering means G2 and 25 that perform filtering processing on the output value from the addition means 17 of the transmission line based on the parameters estimated by the parameter estimation means 22, respectively. An adding means for adding the output value from the adding means 16 and the output value of the filter means 24; 26, an adding means for adding the output value from the adding means 17 of the transmission line and the output value of the filter means 23; and 27, an adding means 25. Multiplication means 28 performs multiplication processing on the output value from the transmission line parameter based on the transmission path parameter, and 28 transmits the output value from the addition means 26 to the output value. Multiplication means for performing multiplication processing based on parameters of the transmission path.
[0025]
Next, the actual operation of FIG. 1 will be described with reference to FIG.
An original signal (desired signal) is generated by the original signal generation means 10, and noise is generated by the noise generation means 11. The original signal passes through the filter means 12 and the filter means 14, and the noise components are added by the addition means 16 and 17. The noise signal passes through the filter means 13 and the filter means 15 and is added to the original signal component by the addition means 16 and 17 (step 9).
[0026]
Here, for simplification, the output value of the original signal generating means 10 is Z-converted and S ₁ (z), the output value of the noise generating means 11 is Z-converted and S ₂ (z) The output value of the adding means 16 is Z-converted and Y ₁ (z), the output value of the adding means 17 is Z-converted and Y ₂ (z) The output value of the adding means 25 is Z-converted and V ₁ (z), the output value of the adding means 26 is Z-converted and V ₂ (z) Z-converts the impulse response of the filter means 12 to H ₁₁ (z) Z-converts the impulse response of the filter means 13 to H ₁₂ (z) Z-converts the impulse response of the filter means 14 to H _{twenty one} (z) Z-converts the impulse response of the filter means 15 to H _{twenty two} (z) Z-convert the impulse response of the filter means 23 to G ₁ (z) Z-converts the impulse response of the filter means 24 to G ₂ Indicated as (z).
[0027]
Further, when each signal is expressed in vector using the above notation, (Expression 2), (Expression 3), and (Expression 4) can be expressed.
[0028]
[Expression 2]

[0029]
[Equation 3]

[0030]
[Expression 4]

[0031]
Y (z) can be expressed by (Equation 5) from (Equation 2), (Equation 3), and (Equation 4).
[0032]
[Equation 5]

[0033]
The Fourier transform means 18 is Y ₁ (z), Y ₂ Perform a short-time Fourier transform process or a Fourier transform process on each of (z), and Y1 (ω), Y ₂ (ω) is calculated (step 10).
[0034]
The cross-correlation calculating means 19 outputs Y1 (ω), Y from the Fourier transform means 18. ₂ A cross-correlation calculation is performed from (ω) (step 11).
[0035]
Cross-correlation calculation output is P _y1y2, P _y2y1 Then, the cross-correlation calculation is expressed by (Expression 6) and (Expression 7).
[0036]
[Formula 6]

[0037]
[Expression 7]

[0038]
The autocorrelation calculating means 20 outputs Y1 (ω), Y from the Fourier transform means 18. ₂ An autocorrelation calculation is performed from (ω) (step 12).
[0039]
Autocorrelation calculation output is P _y1y1 , P _y2y2 Then, the autocorrelation calculation is expressed by (Equation 8) and (Equation 9).
[0040]
[Equation 8]

[0041]
[Equation 9]

[0042]
Neglect means 21 is an autocorrelation calculation output P _y1y1 , P _y2y2 A value below a certain reference value is ignored (step 13). This is necessary in order to stabilize the calculation in the parameter estimation means 22.
[0043]
The parameter estimation means 22 outputs the cross-correlation calculation output P _y1y2, P _y2y1 And Neglected autocorrelation calculation output P _y1y1 , P _y2y2 To G ₁ (ω), G ₂ (ω) is calculated and G ₁ (ω), G ₂ (ω) is inverse Fourier transformed and G ₁ (z), G ₂ (z) is calculated and used as the tap coefficient values of the filter means 23 and the filter means 24, respectively.
[0044]
(Step 14) G ₁ (ω), G ₂ The formula for calculating (ω) is expressed by (Equation 10) and (Equation 11).
[0045]
[Expression 10]

[0046]
[Expression 11]

[0047]
The filter means 23 outputs the output value Y from the adding means 16. ₁ Tap coefficient G for (z) ₁ Filtering processing is performed with (z), and the filter means 24 outputs the output value Y from the adding means 17. ₂ Tap coefficient G for (z) ₂ Filtering processing is performed with (z) (step 15).
[0048]
The adding means 25 adds the output values from the adding means 16 and the filter means 24, and the adding means 26 adds the output values of the adding means 17 and the filter means 23 (step 16).
[0049]
The multiplication means 27 multiplies the output from the addition means 25 by 1 / Δ calculated based on the parameters of the filter means 12, 13, 14, 15 of the transmission line, and the multiplication means 28 multiplies each filter of the transmission line. The output from the adding means 26 is multiplied by 1 / Δ calculated based on the parameters of the means 12, 13, 14, and 15 (step 17), and the process ends (step 18).
[0050]
1 / Δ is calculated by (Equation 12).
[0051]
[Expression 12]

[0052]
According to the present invention, it is possible to provide a method capable of separating an original signal and noise without using a training signal.
[0053]
(Embodiment 2)
FIG. 3 is a functional block diagram of a signal separation method according to the second embodiment of the present invention.
[0054]
In FIG. 3, when a signal is transmitted from the transmission side to the reception side using a twisted pair cable, the signal voltage of the signal line is superimposed on the ground line at an arbitrary amplification factor in the twisted pair cable, and the ground voltage value on the transmission side and the reception side In order to obtain the same ground voltage value as that of the transmitting side on the receiving side, 48 is an original signal generating means for generating an original signal, and 49 is a digital signal. Digital / analog converting means for converting to an analog signal, 50 is a noise generating means for generating an unnecessary signal separately from the original signal, 51 and 53 are arbitrary tap lengths for the original signal output from the original signal generating means, and Filter means H11, H21, 52, 54 for filtering arbitrary tap coefficients are arbitrary tap lengths and arbitrary for the noise signal output from the noise generating means. Filter means H12, H22, and 55 for performing the filter coefficient filtering process add the output of the filter means 51 and the output of the filter means 52, and superimpose the filtered original signal and noise, and 56 denotes the filter means 53. An adding means for adding the outputs of the filter means 54 and superimposing the filtered noise and the original signal, 57 an analog / digital converting means for converting the output analog signal from the adding means 55 into a digital signal, and 58 for adding from the adding means 56 An analog / digital conversion means 59 for converting the output analog signal into a digital signal, 59 is a Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on the output signals from the analog / digital conversion means 57, 58, 60 Performs cross-correlation operation on output value from Fourier transform means 59 Cross-correlation calculating means 61, auto-correlation calculating means 61 for performing auto-correlation calculation on the output value from the Fourier transform means 59, and 62 having an amplitude greater than a certain reference value with respect to the output value from the auto-correlation calculating means Neglect means for ignoring the value, 63 is a twisted pair cable including the filter means 51, the filter means 52, the filter means 53, the filter means 54, and the adding means 55, 56 from the output values of the cross correlation calculating means 60 and the neglect means 62. Parameter estimating means 64 for estimating the parameters, 64 is a filtering means G1 and 65 for filtering the output value from the analog / digital converting means 57 based on the parameters estimated by the parameter estimating means 63, and the parameter estimating means 63 The output value from the analog / digital conversion means 58 based on the parameters estimated by Filter means G2 and 66 for performing filtering processing are added means for adding the output value from the analog / digital conversion means 57 and the output value of the filter means 65, and 67 is an output value from the analog / digital conversion means 58 and the filter means. 64 is an adding means for adding 64 output values, 68 is a multiplying means for multiplying the output value from the adding means 66 based on the twisted pair cable parameters, and 69 is a twisted pair cable for the output value from the adding means 67. Multiplication means for performing a multiplication process based on the parameters.
[0055]
Next, the actual operation of FIG. 3 will be described with reference to FIG.
An original signal (desired signal) is generated by the original signal generating means 48 and converted into an analog signal by the digital / analog converting means 49. Further, noise is generated by the noise generating means 50. The analog original signal passes through the filter means 51 and filter means 53, and the noise components are added by the addition means 55 and 56. The noise signal passes through the filter means 52 and the filter means 54 and is added to the original signal component by the addition means 55 and 56 (step 29). Analog / digital conversion means 57 and 58 convert the analog output signals from the addition means 55 and 56 into digital values, respectively (step 30).
[0056]
Here, for simplification, the output value of the original signal generating means 48 is Z-converted and S ₁ (z) The output value of the noise generating means 50 is Z-converted and S ₂ (z), the output value of the analog / digital conversion means 57 is Z-converted to Y ₁ (z) The output value of the analog / digital conversion means 58 is Z-converted to Y ₂ (z) The output value of the adding means 66 is Z-converted and V ₁ (z), the output value of the adding means 67 is Z-converted and V ₂ (z), the impulse response of the filter means 51 is Z-converted and H ₁₁ (z), the impulse response of the filter means 52 is Z-converted and H ₁₂ (z) The impulse response of the filter means 53 is Z-converted and H _{twenty one} (z), the impulse response of the filter means 54 is Z-converted and H _{twenty two} (z) Z-transform the impulse response of the filter means 64 to obtain G ₁ (z) Z-converts the impulse response of the filter means 65 to G ₂ Indicated as (z).
[0057]
Further, when each signal is expressed in vector using the above notation, (Expression 2), (Expression 3), and (Expression 4) can be expressed.
[0058]
Y (z) can be expressed by (Equation 5) from (Equation 2), (Equation 3), and (Equation 4).
The Fourier transform means 59 is Y ₁ (z), Y ₂ Perform a short-time Fourier transform process or a Fourier transform process on each of (z), and Y1 (ω), Y ₂ (ω) is calculated (step 31).
[0059]
The cross-correlation calculating means 60 outputs Y1 (ω), Y from the Fourier transform means 59. ₂ A cross-correlation calculation is performed from (ω) (step 32).
[0060]
Cross-correlation calculation output is P _y1y2, P _y2y1 Then, the cross-correlation calculation is expressed by (Expression 6) and (Expression 7).
[0061]
Autocorrelation calculating means 61 outputs Y1 (ω), Y from Fourier transform means 59. ₂ An autocorrelation calculation is performed from (ω) (step 33).
[0062]
Autocorrelation calculation output is P _y1y1 , P _y2y2 Then, the autocorrelation calculation is expressed by (Equation 8) and (Equation 9).
[0063]
Neglect means 62 outputs autocorrelation calculation output P _y1y1 , P _y2y2 The value below a certain reference value is ignored (step 34). This is necessary in order to stabilize the calculation in the parameter estimation means 63.
[0064]
The parameter estimation means 63 outputs the cross-correlation calculation output P _y1y2, P _y2y1 And Neglected autocorrelation calculation output P _y1y1 , P _y2y2 To G ₁ (ω), G ₂ (ω) is calculated and G ₁ (ω), G ₂ (ω) is inverse Fourier transformed and G ₁ (z), G ₂ (z) is calculated and used as the tap coefficient values of the filter means 64 and the filter means 65, respectively (step 35). G ₁ (ω), G ₂ The formula for calculating (ω) is expressed by (Equation 10) and (Equation 11).
[0065]
The filter means 64 outputs the output value Y from the analog / digital conversion means 57. ₁ Tap coefficient G for (z) ₁ (z) and the filtering means 65 outputs the output value Y from the analog / digital conversion means 58. ₂ Tap coefficient G for (z) ₂ The filtering process is performed with (z) (step 36).
[0066]
The adding means 66 adds the output values from the analog / digital converting means 57 and the filter means 65, and the adding means 67 adds the output values of the analog / digital converting means 58 and the filter means 64 (step 37).
[0067]
The multiplication means 68 multiplies the output from the addition means 66 by 1 / Δ calculated based on the parameters of the filter means 51, 52, 53, 54 of the twisted pair cable, and the multiplication means 69 multiplies the filters of the twisted pair cable. The output from the adding means 67 is multiplied by 1 / Δ calculated based on the parameters of the means 51, 52, 53, 54 (step 38), and the process is terminated (step 39).
[0068]
1 / Δ is calculated by (Equation 12).
According to the present invention, it is possible to realize a signal separation device that can separate an original signal and noise in communication using a twisted pair cable without using a training signal.
[0069]
(Embodiment 3)
FIG. 5 is a functional block diagram of a signal separation method according to the third embodiment of the present invention.
[0070]
In FIG. 5, in a car navigation system voice recognition device, a mobile phone, OFDM (Orthogonal Frequency Division Multiplexing) communication, etc., the car navigation system voice recognition device receives a car navigation from a speaker (sender). When a voice command is transmitted to the device (reception side), the voice command (original signal) issued by the speaker has an obstacle to the identification of a voice command such as a friction sound with the engine device in the vehicle or the road surface of the tire device. In order to reproduce a voice command buried in the noise in the car navigation device when the noise signal is superimposed with an arbitrary amplification factor and the voice command cannot be identified in the car navigation device, the voice of the speaker is used in the mobile phone device. Ambient noise (noise signal) other than the speaker is arbitrarily amplified with respect to the signal (original signal) In order to reproduce the voice (original signal) of the speaker buried in noise in the mobile phone device on the transmission side when the speaker's caller cannot recognize the voice, the multiple transmission is performed in OFDM communication. The present invention is applied to a signal transmitted from a station (the original signal is taken out at the receiving station and only the signal (original signal) of the desired station is applied. In the figure, reference numeral 70 denotes an original signal generating means for generating the original signal. 71, noise generating means for generating an unnecessary signal separately from the original signal, 72, 74 filter means for performing filtering processing of arbitrary tap length and arbitrary tap coefficient on the original signal output from the original signal generating means H11, H21, 73, and 75 are filter means H12, H22, and 76 that perform filtering of an arbitrary tap length and an arbitrary tap coefficient on the noise signal output from the noise generating means. An adding means for adding the output of the stage 72 and the output of the filter means 73 and superimposing the filtered original signal and the noise, 77 adds the outputs of the filter means 74 and the filter means 75, and the filtered noise and the original signal are added. An adding unit 78 for superimposing, a microphone for converting a signal from the adding unit 76 into an electric signal, a microphone for converting a signal from the adding unit 77 into an electric signal, and an analog electric signal output from the microphone 78 as a digital signal. An analog / digital conversion means 81 for converting the output signal from the microphone 79 into an analog / digital conversion means for converting an analog electrical signal output from the microphone 79 into a digital signal, and 82 for a short time with respect to the output signals from the analog / digital conversion means 80, 81. A Fourier transform process or a Fourier transform process Rie transforming means, 83 is a cross-correlation calculating means for performing a cross-correlation calculation on the output value from the Fourier transforming means 82, 84 is an autocorrelation calculating means for performing an autocorrelation calculation on the output value from the Fourier transforming means 82, Reference numeral 85 is neglect means for ignoring a value greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculation means, and reference numeral 86 is a filter means 72, filter based on the output values of the cross correlation calculation means 83 and neglect means 85. Parameter estimating means 87 for estimating transmission path parameters including means 73, filter means 74, filter means 75, and addition means 76 and 77, and analog / digital conversion means 80 based on the parameters estimated by parameter estimation means 86. The filter means G 1 and 88 for performing the filtering process on the output value from are similarly estimated by the parameter estimation means 86. Filter means G2 and 89 for performing filtering processing on the output value from the analog / digital conversion means 81 based on the parameters thus obtained are additions for adding the output value from the analog / digital conversion means 80 and the output value of the filter means 88. Means 90 is an addition means for adding the output value from the analog / digital conversion means 81 and the output value of the filter means 87, and 91 is a multiplication process for the output value from the addition means 89 based on the parameters of the transmission path. Multiplication means 92 is a multiplication means for performing multiplication processing on the output value from the addition means 90 based on the parameters of the transmission path.
[0071]
Next, the actual operation of FIG. 5 will be described with reference to FIG.
An original signal (desired signal) is generated by the original signal generating means 70, and noise is generated by the noise generating means 71. The original signal passes through the filter means 72 and filter means 74, and the noise components are added by the addition means 76 and 77. The noise signal passes through the filter means 73 and the filter means 75 and is added to the original signal component by the addition means 76 and 77 (step 40). The microphones 78 and 79 convert the audio signals into electric signals for the audio signals from the adding means 76 and 77 (step 41). Analog / digital conversion means 80 and 81 convert analog electric signals from the microphones 78 and 79 into digital values, respectively (step 42).
[0072]
Here, for simplification, the output value of the original signal generating means 70 is Z-converted and S ₁ (z), the output value of the noise generating means 71 is Z-converted and S ₂ (z) Z-converts the output value of the analog / digital conversion means 80 to Y ₁ (z) Z-convert the output value of the analog / digital conversion means 81 to Y ₂ (z), the output value of the adding means 89 is Z-converted and V ₁ (z), the output value of the adding means 90 is Z-converted and V ₂ (z), the impulse response of the filter means 72 is Z-converted and H ₁₁ (z) The impulse response of the filter means 73 is Z-converted and H ₁₂ (z) The impulse response of the filter means 74 is Z-converted and H _{twenty one} (z) The impulse response of the filter means 75 is Z-converted and H _{twenty two} (z) Z-converts the impulse response of the filter means 87 to G ₁ (z), the impulse response of the filter means 88 is Z-converted and G ₂ Indicated as (z).
[0073]
Further, when each signal is expressed in vector using the above notation, (Expression 2), (Expression 3), and (Expression 4) can be expressed.
[0074]
Y (z) can be expressed by (Equation 5) from (Equation 2), (Equation 3), and (Equation 4).
The Fourier transform means 82 is Y ₁ (z), Y ₂ Perform a short-time Fourier transform process or a Fourier transform process on each of (z), and Y1 (ω), Y ₂ (ω) is calculated (step 43).
[0075]
The cross-correlation calculating means 83 outputs Y1 (ω), Y from the Fourier transform means 82. ₂ A cross-correlation calculation is performed from (ω) (step 44).
[0076]
Cross-correlation calculation output is P _y1y2, P _y2y1 Then, the cross-correlation calculation is expressed by (Expression 6) and (Expression 7).
[0077]
The autocorrelation calculating means 84 outputs Y1 (ω), Y from the Fourier transform means 82. ₂ An autocorrelation calculation is performed from (ω) (step 45).
[0078]
Autocorrelation calculation output is P _y1y1 , P _y2y2 Then, the autocorrelation calculation is expressed by (Equation 8) and (Equation 9).
[0079]
Neglect means 85 outputs autocorrelation calculation output P _y1y1 , P _y2y2 The value below a certain reference value is ignored (step 46). This is necessary in order to stabilize the calculation in the parameter estimation means 86.
[0080]
The parameter estimation means 86 outputs the cross-correlation calculation output P _y1y2, P _y2y1 And Neglected autocorrelation calculation output P _y1y1 , P _y2y2 To G ₁ (ω), G ₂ (ω) is calculated and G ₁ (ω), G ₂ (ω) is inverse Fourier transformed and G ₁ (z), G ₂ (z) is calculated and used as the tap coefficient values of the filter means 87 and the filter means 88, respectively. (Step 47) G ₁ (ω), G ₂ The formula for calculating (ω) is expressed by (Equation 10) and (Equation 11).
[0081]
The filter means 87 outputs the output value Y from the analog / digital conversion means 80. ₁ Tap coefficient G for (z) ₁ Filtering processing is performed with (z), and the filter means 88 outputs the output value Y from the analog / digital conversion means 81. ₂ Tap coefficient G for (z) ₂ Filtering processing is performed with (z) (step 48).
[0082]
The adding means 89 adds the output values from the analog / digital converting means 80 and the filter means 88, and the adding means 90 adds the output values of the analog / digital converting means 81 and the filter means 87 (step 49).
[0083]
The multiplication means 91 multiplies the output from the addition means 89 by 1 / Δ calculated based on the parameters of the filter means 72, 73, 74, 75 of the transmission path, and the multiplication means 92 multiplies each filter of the transmission path. The output from the adding means 90 is multiplied by 1 / Δ calculated based on the parameters of the means 72, 73, 74, 75 (step 50), and the process is terminated (step 51).
[0084]
1 / Δ is calculated by (Equation 12).
According to the present invention, there is provided a signal separation device capable of removing only a desired signal by removing ambient noise (signals of other channels in OFDM communication) in voice recognition means of a car navigation system, a portable telephone, OFDM communication, etc. without using a training signal. realizable.
[0085]
(Embodiment 4)
FIG. 7 is a functional block diagram of a signal separation method according to the fourth embodiment of the present invention.
[0086]
93 is a short-time Fourier transform process for the input signal, or a Fourier transform means for performing a Fourier transform process, 94 is a cross-correlation operation means for performing a cross-correlation operation from an output signal from the Fourier transform means 93, and 95 is a Fourier transform means 93. Autocorrelation calculating means for performing an autocorrelation calculation from the output signal from, 96 is a neglecting means for ignoring a value below a certain reference value with respect to the output signal from the autocorrelation calculating means 95, and 97 is a cross correlation calculating means 94 This is parameter estimation means for estimating the parameters of the transmission line from the output signal and the output signal from the neglect means 96.
[0087]
Next, the actual operation of FIG. 7 will be described with reference to FIG.
When a signal is input to the Fourier transform means 93 (step 52), the Fourier transform means 93 performs a short-time Fourier transform operation or a Fourier transform operation on each of the two systems of input signals (step 53). The cross-correlation calculation means performs cross-correlation calculation on the output signal from the Fourier transform means (step 54), and the auto-correlation calculation means similarly performs auto-correlation calculation (step 55). Since the calculation takes time for Step 54 and Step 55, independent means are assembled as hardware, and Step 54 and Step 55 perform the calculation in parallel.
[0088]
The Neglect unit 96 ignores a value less than a certain reference value with respect to the output signal from the autocorrelation calculation unit 95 (step 56), and the parameter estimation unit uses the output signal from the cross-correlation calculation unit 94 and the Neglect unit 96. The calculation shown in (Equation 10) and (Equation 11) is performed from the output signal of (5), the inverse Fourier transform of the result is performed (step 57), and the process is terminated (step 58).
[0089]
【The invention's effect】
As described above, according to the present invention, since the original signal (desired signal) and noise can be separated without using a training signal, the throughput is improved in communication, and the transmission path is time-consuming. Even in the case of a continuous change, the original signal (desired signal) and noise can be separated following the change. In addition, since the car navigation device can receive voice commands from which in-vehicle noise has been removed, the voice recognition rate is improved, and the cellular phone device can remove noise around the speaker, enabling high-quality calls without noise. In OFDM communication, there is an effect that a signal of only a desired station can be received with high quality.
[Brief description of the drawings]
FIG. 1 is a functional block diagram of a signal separation method according to a first embodiment of the present invention.
FIG. 2 is a flowchart showing the operation of the signal separation method according to the first embodiment of the present invention.
FIG. 3 is a functional block diagram of a signal separation device according to a second embodiment of the present invention.
FIG. 4 is a flowchart showing an operation of the signal separation device according to the second exemplary embodiment of the present invention.
FIG. 5 is a functional block diagram of a signal separation device according to a third embodiment of the present invention.
FIG. 6 is a flowchart showing the operation of the signal separation device according to the third embodiment of the present invention.
FIG. 7 is a functional block diagram of a signal separation device according to a fourth embodiment of the present invention.
FIG. 8 is a flowchart showing the operation of the signal separation device according to the fourth exemplary embodiment of the present invention.
FIG. 9 is a functional block diagram of a conventional signal separation device.
FIG. 10 is a flowchart showing the operation of a conventional signal separation device.
[Explanation of symbols]
1 Original signal generation means
2 Training signal storage means
3 Switching means
4 Noise generation means
5 addition means
6 Transmission path filter means
7 Filter processing means
8 Training signal storage means
9 Transmission path inverse characteristic calculation means
10 Original signal generation means
11 Noise generation means
12 Filter means
13 Filter means
14 Filter means
15 Filter means
16 Addition means
17 Adding means
18 Fourier transform means
19 Cross-correlation calculation means
20 Autocorrelation calculation means
21 Neglect means
22 Parameter estimation means
23 Filter means
24 Filter means
25 addition means
26 Adding means
27 Multiplication means
28 Multiplying means
48 Original signal generating means
49 Digital / analog conversion means
50 Noise generation means
51 Filter means
52 Filter means
53 Filter means
54 Filter means
55 Adding means
56 Adding means
57 Analog / digital conversion means
58 Analog / digital conversion means
59 Fourier transform means
60 Cross-correlation calculation means
61 Autocorrelation calculation means
62 Neglect means
63 Parameter estimation means
64 Filter means
65 Filter means
66 Adding means
67 Adding means
68 Multiplying means
69 Multiplication means
70 Original signal generating means
71 Noise generation means
72 Filter means
73 Filter means
74 Filter means
75 Filter means
76 Adding means
77 Adding means
78 microphone
79 Microphone
80 Analog / digital conversion means
81 Analog / digital conversion means
82 Fourier transform means
83 Cross-correlation calculation means
84 Autocorrelation calculation means
85 Neglect means
86 Parameter estimation means
87 Filter means
88 Filter means
89 Adding means
90 addition means
91 Multiplication means
92 Multiplying means
93 Fourier transform means
94 Cross-correlation calculation means
95 Autocorrelation calculation means
96 Neglect means
97 Parameter estimation means

Claims (8)

First and second filter means for filtering arbitrary tap lengths and arbitrary tap coefficients on the original signal output from the original signal generating means;
Third and fourth filter means for filtering arbitrary tap lengths and arbitrary tap coefficients on the noise signal output from the noise generating means;
First addition means for adding the output of the first filter means and the output of the third filter means, and superimposing the filtered original signal and noise;
Second addition means for adding the outputs of the second filter means and the fourth filter means, and superimposing the filtered noise and the original signal;
Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on the output signal from the first addition means;
Cross-correlation calculation means for performing cross-correlation calculation on the output value from the Fourier transform means;
Autocorrelation calculation means for performing autocorrelation calculation on the output value from the Fourier transform means;
Neglect means for ignoring the value above a certain reference value with respect to the output value from the autocorrelation calculating means,
Parameter estimation means for estimating the parameters of the transmission line including the first to fourth filter means and the first and second addition means from the output values of the cross-correlation calculation means and the neglect means;
A fifth filter means for performing a filtering process on the output value from the first addition means of the transmission line based on the parameter estimated by the parameter estimation means;
Similarly, a sixth filter means for performing a filtering process on the output value from the second addition means of the transmission line based on the parameter estimated by the parameter estimation means,
Third addition means for adding the output value from the first addition means of the transmission line and the output value of the sixth filter means;
A fourth addition means for adding the output value from the second addition means of the transmission line and the output value of the fifth filter means;
First multiplication means for performing a multiplication process on the output value from the third addition means based on the parameters of the transmission path;
A signal separation device comprising: second multiplication means for performing multiplication processing on the output value from the fourth addition means based on a transmission path parameter. In communication using twisted pair cable,
A digital / analog converting means for converting a digital output signal from the original signal generating means into an analog signal;
First and second filter means for performing filtering processing of an arbitrary tap length and an arbitrary tap coefficient on a signal output from the digital / analog converting means;
Third and fourth filter means for filtering arbitrary tap lengths and arbitrary tap coefficients on the noise signal output from the noise generating means;
First addition means for adding the output of the first filter means and the output of the third filter means, and superimposing the filtered analog original signal and noise;
Second addition means for adding the outputs of the second filter means and the fourth filter means, and superimposing the filtered noise and the analog original signal;
First analog / digital conversion means for converting an analog output signal from the first addition means into a digital signal;
Second analog / digital conversion means for converting an analog output signal from the second addition means to a digital signal;
Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on the output signals from the first and second analog / digital conversion means;
Cross-correlation calculation means for performing cross-correlation calculation on the output value from the Fourier transform means;
Autocorrelation calculation means for performing autocorrelation calculation on the output value from the Fourier transform means;
Neglect means for ignoring the value above a certain reference value with respect to the output value from the autocorrelation calculating means,
Parameter estimation means for estimating the parameters of the transmission line including the first to fourth filter means and the first and second addition means from the output values of the cross-correlation calculation means and the neglect means;
Fifth filter means for performing a filtering process on the output value from the first addition means of the twisted pair cable based on the parameter estimated by the parameter estimation means;
A sixth filter means for performing a filtering process on the output value from the second addition means of the twisted pair cable based on the parameter similarly estimated by the parameter estimation means;
Third addition means for adding the output value from the addition means of the twisted pair cable and the output value of the sixth filter means;
Fourth addition means for adding the output value from the second addition means of the twisted pair cable and the output value of the fifth filter means;
First multiplication means for performing multiplication processing on the output value from the third addition means based on the parameters of the twisted pair cable;
A signal separation device comprising second multiplication means for performing multiplication processing on an output value from the fourth addition means based on a twisted pair cable parameter. In a car navigation system voice recognition device, mobile phone, OFDM communication, etc., it extracts only a desired signal from a desired signal mixed with noise (signal of another channel in OFDM communication),
First and second filter means for filtering arbitrary tap lengths and arbitrary tap coefficients with respect to the signal output from the original signal generating means;
Third and fourth filter means for filtering arbitrary tap lengths and arbitrary tap coefficients on the noise signal output from the noise generating means;
First addition means for adding the output of the first filter means and the output of the third filter means, and superimposing the filtered original signal and noise;
Second addition means for adding the outputs of the second filter means and the fourth filter means, and superimposing the filtered noise and the original signal;
A first microphone for converting an output signal from the first adding means into an electrical signal;
A second microphone for converting the output signal from the second adding means into an electrical signal;
First analog / digital conversion means for converting an analog output signal from the first microphone into a digital signal;
Second analog / digital conversion means for converting an analog output signal from the second microphone into a digital signal;
Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on the output signals from the first and second analog / digital conversion means;
Cross-correlation calculation means for performing cross-correlation calculation on the output value from the Fourier transform means;
Autocorrelation calculation means for performing autocorrelation calculation on the output value from the Fourier transform means;
Neglect means for ignoring the value above a certain reference value with respect to the output value from the autocorrelation calculating means,
Parameter estimation means for estimating the parameters of the transmission line including the first to fourth filter means and the first and second addition means from the output values of the cross-correlation calculation means and the neglect means;
A fifth filter means for performing a filtering process on the output value from the first addition means of the transmission line based on the parameter estimated by the parameter estimation means;
Similarly, a sixth filter means for performing a filtering process on the output value from the second addition means of the transmission line based on the parameter estimated by the parameter estimation means,
Third addition means for adding the output value from the first addition means of the transmission line and the output value of the sixth filter means;
A fourth addition means for adding the output value from the second addition means of the transmission line and the output value of the fifth filter means;
First multiplication means for performing a multiplication process on the output value from the third addition means based on the parameters of the transmission path;
A signal separation device comprising: second multiplication means for performing multiplication processing on the output value from the fourth addition means based on a transmission path parameter. Fourier transform means for performing short-time Fourier transform processing or Fourier transform processing on two input signals;
Cross-correlation calculation means for performing cross-correlation calculation on the output value from the Fourier transform means;
Autocorrelation calculation means for performing autocorrelation calculation on the output value from the Fourier transform means;
Neglect means for ignoring the value above a certain reference value with respect to the output value from the autocorrelation calculating means,
Signals characterized by having parameter estimation means for estimating transmission path parameters from the output values of the cross-correlation calculation means and neglect means, and performing high-speed calculation by parallel processing of the cross-correlation calculation means and the autocorrelation calculation means Separation device. Filtering the arbitrary tap length and the arbitrary tap coefficient by the first and second filter means on the original signal output from the original signal generating means;
Filtering the arbitrary tap length and arbitrary tap coefficient by the third and fourth filter means on the noise signal output from the noise generating means;
A step of adding the output of the first filter means and the output of the third filter means by the first addition means, and superimposing the filtered original signal and noise; a second filter means and a fourth filter means; Are added by the second adding means,
Superimposing the filtered noise and the original signal;
Performing a short-time Fourier transform process or a Fourier transform process on the output signals from the first addition means and the second addition means in the Fourier transform means;
Performing a cross-correlation operation on the output value from the Fourier transform unit by the cross-correlation calculation unit;
Performing an autocorrelation operation on the output value from the Fourier transform means by the autocorrelation operation means;
A step of ignoring by the neglect means for a value greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculating means,
Estimating the parameters of the transmission line including the first to fourth filter means and the first and second addition means from the output values of the cross-correlation calculation means and the neglect means by the parameter estimation means;
Filtering the output value from the first addition means of the transmission line by the fifth filter means based on the parameter estimated by the parameter estimation means;
Filtering the output value from the second addition means of the transmission line on the basis of the parameter similarly estimated by the parameter estimation means by the sixth filter means;
Adding the output value from the first addition means of the transmission line and the output value of the sixth filter means by the third addition means;
Adding the output value from the second addition means of the transmission line and the output value of the fifth filter means by the third addition means;
A step of multiplying the output value from the third adding means by the first multiplying means based on the parameters of the transmission path;
And a step of multiplying the output value from the fourth adding means by the second multiplying means based on the parameters of the transmission line. Performing a short-time Fourier transform process or a Fourier transform process on the two input signals by a Fourier transform means;
Performing a cross-correlation operation on the output value from the Fourier transform unit by the cross-correlation calculation unit;
Performing an autocorrelation calculation on the output value from the Fourier transform means by the autocorrelation calculation means;
A step of ignoring by the neglect means for a value greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculating means,
A step of estimating the parameter of the transmission path by the parameter estimating means from the output value of the cross correlation calculating means and the neglect means,
A signal separation method characterized by performing high-speed computation by processing a cross-correlation computing means and an autocorrelation computing means in parallel. A recording medium recording a signal separation method for extracting only an original signal from a signal in which an original signal and a noise signal are superimposed,
Filtering the arbitrary tap length and the arbitrary tap coefficient by the first and second filter means on the original signal output from the original signal generating means;
Filtering the arbitrary tap length and arbitrary tap coefficient by the third and fourth filter means on the noise signal output from the noise generating means;
Adding the output of the first filter means and the output of the third filter means by the first addition means, and superimposing the filtered original signal and noise;
Adding the outputs of the second filter means and the fourth filter means by the second addition means, and superimposing the filtered noise and the original signal;
Performing a short-time Fourier transform process or a Fourier transform process on the output signals from the first addition means and the second addition means in the Fourier transform means;
Performing a cross-correlation operation on the output value from the Fourier transform unit by the cross-correlation calculation unit;
Performing an autocorrelation operation on the output value from the Fourier transform means by the autocorrelation operation means;
A step of ignoring by the neglect means for a value greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculating means,
Estimating the parameters of the transmission line including the first to fourth filter means and the first and second addition means from the output values of the cross-correlation calculation means and the neglect means by the parameter estimation means;
Filtering the output value from the first addition means of the transmission line by the fifth filter means based on the parameter estimated by the parameter estimation means;
Filtering the output value from the second addition means of the transmission line on the basis of the parameter similarly estimated by the parameter estimation means by the sixth filter means;
Adding the output value from the first addition means of the transmission line and the output value of the sixth filter means by the third addition means;
Adding the output value from the second addition means of the transmission line and the output value of the fifth filter means by the third addition means;
A step of multiplying the output value from the third adding means by the first multiplying means based on the parameters of the transmission path;
A recording medium having recorded thereon a program for executing a step of performing a multiplication process by a second multiplication unit on the basis of a transmission path parameter with respect to an output value from a fourth addition unit. Performing a short-time Fourier transform process or a Fourier transform process on the two input signals by a Fourier transform means;
Performing a cross-correlation operation on the output value from the Fourier transform unit by the cross-correlation calculation unit;
Performing an autocorrelation calculation on the output value from the Fourier transform means by the autocorrelation calculation means;
A step of ignoring by the neglect means for a value greater than a certain reference value with an amplitude relative to the output value from the autocorrelation calculating means,
A recording medium on which a program for executing a step of estimating a parameter of a transmission path by a parameter estimation unit from output values of a cross-correlation calculation unit and a neglect unit is recorded.

Images