JP3728837B2 - Active noise control device - Google Patents

Description

【００６９】
である。そして、行列の各成分の自乗の総和を‖‖² という記号で表すと、式(41)の右辺は‖ＡＺ−Ｂ‖² と表わされる。
行列Ａの転置行列を ^TＡと表せば、Ａの一般逆行列は（ ^TＡＡ）^{-1 T}Ａであり、
Ｚ＝（ ^TＡＡ）^{-1 T}ＡＢ (42)
が‖ＡＺ−Ｂ‖² を最小にするという一般逆行列の性質は周知の通りである（例えば、伊藤昇、岩井齊良、岩堀長慶、上林達治、関野薫、高橋秀一 「経済系・工学系のための 行列とその応用」、紀伊國屋書店、pp. ７１−７２（1987）参照）。
【００７０】
未知数Ｚを求めれば、これがすなわち消音フィルタ７に設定すべき係数列ｈを表している。
なお、式(39)、式(40)のために周波数領域で表したＨが必要ならば、ｈに対してＦＦＴを施せば算出できる。
計算の効率化のために、式(42)を直接計算する代わりに、
（ ^TＡＡ）Ｚ＝ ^TＡＢ (43)
なる連立方程式を解くことでＺを求めることもできる。ｃの自己相関関数をｒ_i とすると、左辺の（ ^TＡＡ）のｉ行ｊ列の要素はｉ≧ｊのときｒ_i-j 、ｉ＜ｊのときｒ_j-i である。（ ^TＡＡ）は、対称Toeplitz行列（左上から右下ヘの斜め方向に並ぶ要素が同一の対称行列）となるので、Levinsonアルゴリズムが使える。
【００７１】
また、右辺の ^TＡＢは、ｃと−ｇの相互相関関数になっている。
ｃの自己相関関数は ^*ＣＣのＩＦＦＴで効率よく計算してもよい（ ^*ＣはＣの共役複素数）。
また、ｃとｇの相互相関関数は、 ^*ＣＧのＩＦＦＴで効率よく計算してもよい。いうまでもなく、ＩＦＦＴを使うとインパルス応答ｃやｇを一旦求める必要がない。後述の§５−２と§５−３はこの性質を利用している。
【００７２】
なお、最初に求まっているのがインパルス応答ｇとｃである場合に、消音フィルタ係数ｈを式(43)で算出すること自体は、例えば、鈴木陽一、浅野太、曽根敏夫、「音響系の伝搬系の模擬をめぐって（その２）」日本音響学会誌、４５巻１号、p.４４（1989）などに記載されている。
また、消音フィルタ係数ｈの算出に関連する公知の改良手法、例えば、^TＡＡに対角要素σＩを加算しておく（σは定数、Ｉは単位行列）ことで、^TＡＡの行列式が小さな値になる場合に、係数ｈの値が異常に大きくなることを防げること（大賀寿郎、山崎芳男、金田豊、「音響システムとディジタル処理」、電子情報通信学会、p. 155（1995）参照）が本発明にも利用できるのはいうまでもない。
【００７３】
§５−２．ΣＷ＜｜Ｘ｜²＞｜Ｇ＋ＨＣ｜²を最小にする消音フィルタ
次に、式(36)に出てくる重み付けＷ＜｜Ｘ｜²＞を含めた
ΣＷ＜｜Ｘ｜²＞｜Ｇ＋ＨＣ｜²
を最小にする消音フィルタの計算方法を説明する。
いま、Ｗ＜｜Ｘ｜²＞＝Ｍ² とおくとする（Ｍは負でない実数）。すると、
Ｗ＜｜Ｘ｜²＞｜Ｇ＋ＨＣ｜² ＝｜（ＭＧ）＋Ｈ（ＭＣ）｜²
となり、
^*（ＭＣ）（ＭＣ）＝^*ＣＣＭ² ，
^*（ＭＣ）（ＭＧ）＝^*ＣＧＭ²
が成り立つ。従って、Ｗ＜｜Ｘ｜²＞なる重み付けを算入するには、§５−１の最後で説明したＩＦＦＴする前の^*ＣＣと^*ＣＧにこの重み付けを乗じて、それぞれ^*ＣＣＷ＜｜Ｘ｜²＞と^*ＣＧＷ＜｜Ｘ｜²＞にしておけばよい。
【００７４】
なお、いうまでもなく、Ｗか＜｜Ｘ｜²＞の片方だけの重み付けを算入するなら、それぞれ、^*ＣＣＷと^*ＣＧＷ、^*ＣＣ＜｜Ｘ｜²＞と^*ＣＧ＜｜Ｘ｜²＞をＩＦＦＴすればよい。
§５−３．ΣＷ＜｜Ｘ｜²＞｜Ｇ＋ＨＬ_IＬ_OＣ｜²を最小にする消音フィルタ
そして、式(36)の右辺にあるΣＷ＜｜Ｘ｜²＞｜Ｇ＋ＨＬ_IＬ_OＣ｜²を最小にする消音フィルタ係数ｈを計算するには、前記のＣの代わりに、Ｌ_IＬ_OＣを与えればよい。このようにして、消音フィルタ係数ｈを求める手順を、以下では、「消音フィルタ算出手順」と呼ぶことにする。
【００７５】
§６．能動騒音制御装置の立ち上げ手順
次に、式(39)、式(40)を使って騒音波伝搬系ＧとＬ_I Ｌ_O Ｃを求める場合の、能動騒音制御装置の立ち上げ処理について説明する。
初期状態として、騒音波伝搬系Ｇと消音波伝搬系Ｃに関する予備知識がない場合には、Ｈ₁ とＨ₂ の２通り（Ｈ₁ ≠Ｈ₂ ）についてＲ₁ とＲ₂ を求めなければ、式(39)、式(40)が計算できない。Ｈ₁ とＨ₂ の片方（単純化のためＨ₁ とする）は０、すなわち、消音波を出さないとすることもできるが、残る他方（Ｈ₂ ）は、とにかく消音波を出さざるを得ない。
【００７６】
そこで、例えば、うるさくなるが許容できる程度の大きさの消音波を出すことにし、かかる大きさａのインパルス波形を係数ｈ₂ に割り当てるとよい。大きさａは例えば１（消音フィルタの入力をそのまま出力する）とか、 0.1（消音フィルタの入力の 0.1倍の信号を消音フィルタから出力する）などを割り当てる。Ｈ₂ は係数ｈ₂ にＦＦＴを施すことによって求める。Ｈ₁ とＨ₂ は、他の値にすることもできるが、Ｈ₁ ＝０、Ｈ₂ ＝ａの場合は、式(39)、式(40)に出てくる除算の計算が単純にできる。式(39)、式(40)によって騒音波伝搬系ＧとＬ_I Ｌ_O Ｃが求まった後は、前述の消音フィルタ算出手順により消音フィルタ係数ｈを求め、これを消音フィルタ７に設定すれば、消音の目標である式(36)を最小にすることができる。
【００７７】
§７_. 消音の維持
§７−１．系のずれの検出
次に、消音動作中に騒音波伝搬系Ｇや消音波伝搬系Ｃが変化する場合に、この変化に追従して消音を維持する手順について説明する。そのために、Ｒを常に観測しておき、Ｌ_I Ｌ_O Ｃ，Ｇ，ｈ，Ｈの値を常に更新するとする。
初期値をＬ_I Ｌ_O Ｃ₁ ，Ｇ₁ ，ｈ₁ ，Ｈ₁ とする（便宜上添字を１から始めるが、このＨ₁ は立ち上げ処理で用いるＨ₁ ＝０とは別物）。系のずれがないときでも常に更新を行うのでもよい。ここでは、無駄な更新を省く観点から、系のずれを検出して、ずれが大きい場合のみ更新を行う場合を示す。手順を以下に示す。
（ｉ）初期状態として消音フィルタがＨ₁ のとき（係数はｈ₁ ）の、Ｒ₁ を求める。
【００７８】
系にずれがないならば、Ｒ₁ はＧ₁ ＋Ｈ₁ Ｌ_I Ｌ_O Ｃ₁ に等しいはずである。。ずれが予め定めた閾値よりも大きければ§７−２の（ii）に進む。系のずれは、消音のずれでもある。ずれの大ききは、例えばΣ｜Ｒ₁ −（Ｇ₁ ＋Ｈ₁ Ｌ_I Ｌ_O Ｃ₁ ）｜² の値の大きさで判断する。
§７−２．系のずれに対処する消音フィルタの変化
（ii）消音フィルタの係数をｈ₂ に変えてみて、Ｒ₂ を求める。
【００７９】
Ｒ₁ −（Ｇ₁ ＋Ｈ₁ Ｌ_I Ｌ_O Ｃ₁ ）は、消音のずれを示す。騒音波伝搬系Ｇと消音波伝搬系Ｃのどちらがずれているかは、Ｒ₁ だけからは決められない。どちらがずれやすいかは、音響系の構成に依存する。従って、どちらが変動しやすいかを決めておき、変動しにくい方を元の値のままとし、変動しやすい方をＲ₁ から仮に求める。
【００８０】
消音により役立つこととなる係数ｈ₂ の値として、以下の▲１▼〜▲３▼の３通りを示しておく。
▲１▼騒音波伝搬系Ｇがずれやすいと考えるなら、
Ｇ＝Ｒ₁ −Ｈ₁ Ｌ_I Ｌ_O Ｃ₁ (44)
とする。消音フィルタの係数ｈ₂ は、この騒音波伝搬系ＧとＣ₁ を使って消音フィルタ算出手順に従って求める。Ｈ₂ はＦＦＴによりｈ₂ から求めておく。
▲２▼消音波伝搬系Ｃがずれやすいと考えるなら、
Ｌ_I Ｌ_O Ｃ＝（Ｒ₁ −Ｇ₁ ）／Ｈ₁ (45)
とする。消音フィルタの係数ｈ₂ は、このＬ_I Ｌ_O Ｃと騒音波伝搬系Ｇ₁ を使って消音フィルタ算出手順に従って求める。Ｈ₂ はＦＦＴによりｈ₂ から求めておく。
▲３▼一方、Ｈ₂ −Ｈ₁ ＝０となる周波数が存在するとその周波数においては、前回同定した値をそのまま参照せざるを得ない。かかる観点からは、Ｈ₂ −Ｈ₁ ＝０となる周波数が存在しない方が望ましいので、簡便に計算できることも勘案して、インパルス波形（全周波数範囲にわたって、パワーを有する）をｈ₁ に付け加えることで、ｈ₂ を実現する場合も説明する。インパルス波形を付け加えるのは、Ｒ₁ とＬ_I Ｌ_O Ｃ₁ の相互相関関数（Ｗ＜｜Ｘ｜² ＞なる重み付き）の絶対値が最大になる時点とし、その大きさは

とする。これによりｈ₂ が求まればそれにＦＦＴを施すことでＨ₂ を求める。
【００８１】
なお、§５−３の説明に従って、式(46)の分子に出てくる「Ｒ₁とＬ_IＬ_OＣ₁ の相互相関関数（Ｗ＜｜Ｘ｜²＞なる重み付き）」は、^*（Ｌ_IＬ_OＣ₁）ＲＷ＜｜Ｘ｜² ＞をＩＦＦＴすることで求まり、式(46)の分母に出てくる「Ｌ_IＬ_OＣ₁の自己相関関数（Ｗ＜｜Ｘ｜²＞なる重み付き）」は、^*（Ｌ_IＬ_OＣ₁）（Ｌ_IＬ_OＣ₁）Ｗ＜｜Ｘ｜²＞をＩＦＦＴすることで求まる。
【００８２】
また、式(46)の分母の「Ｌ_I Ｌ_O Ｃ₁ の自己相関関数（Ｗ＜｜Ｘ｜² ＞なる重み付き）で時刻０の値」は、Ｗ＜｜Ｘ｜² ＞｜Ｌ_I Ｌ_O Ｃ₁ ｜² を全周波数範囲にわたって平均した値に等しいので、これにより計算してもよい。
【００８３】
§７−３．系の再同定
（iii ）Ｇ₁ とＣ₁ をＧ₂ とＣ₂ に更新する。
まず、（ii）のＨ₂ に対するＲ₂ を求める。そして、Ｒ₁ ＝Ｇ＋Ｈ₀ Ｌ_I Ｌ_O ＣとＲ₂ ＝Ｇ＋Ｈ₁ Ｌ_I Ｌ_O Ｃから、式(39)、式(40)を使って現在のＧとＣを計算する。
次に、各周波数におけるΔＨ＝｜Ｈ₂ −Ｈ₁ ｜² の大きさに応じてＧとＣをＧ₁ とＣ₁ に算入し、Ｇ₂ とＣ₂ に更新する。この更新の方法として、以下に▲１▼と▲２▼の二通りを示す。
【００８４】
▲１▼定性的にいって、ΔＨが大きければ、新しく求めたＧとＣは信頼できる。ΔＨ＝０なら更新しない。ΔＨの多寡を峻別するために予め定めた定数をαとし、ΔＨ≧αなら全更新する。０＜ΔＨ＜αなら、ΔＨに比例した割合で更新することにする。つまり、引数の中で小さい方を選ぶ関数ｍｉｎ（ ，）を使って、ｒ＝ｍｉｎ（ΔＨ／α，１）なる値を算出し、
Ｇ₂ ＝（１−ｒ）Ｇ₁ ＋ｒＧ (47)
Ｌ_I Ｌ_O Ｃ₂ ＝（１−ｒ）Ｌ_I Ｌ_O Ｃ₁ ＋ｒＬ_I Ｌ_O Ｃ (48)
で更新する。
【００８５】
なお、０≦ΔＨ≦αの範囲では、

となり、除算を避けることができる。
【００８６】
また、ΔＨ≧αの範囲では、
Ｇ₂ ＝ ^*（Ｈ₂ −Ｈ₁ ）（Ｈ₂ Ｒ₁ −Ｈ₁ Ｒ₂ ）／｜Ｈ₂ −Ｈ₁ ｜² (51)
Ｌ_I Ｌ_O Ｃ₂ ＝ ^*（Ｈ₂ −Ｈ₁ ）（Ｒ₂ −Ｒ₁ ）／｜Ｈ₂ −Ｈ₁ ｜² (52)
となる。従って、実際の計算においては、一旦ＧとＣを中間結果として算出するのではなく、 ^*（Ｈ₂ −Ｈ₁ ）（Ｈ₂ Ｒ₁ −Ｈ₁ Ｒ₂ ）と ^*（Ｈ₂ −Ｈ₁ ）（Ｒ₂ −Ｒ₁ ）を算出しておくのが、適切である。
【００８７】
▲２▼Ｈ₂ −Ｈ₁ で割る除算が生じるが、別のやり方も示しておく。これは、§７−１（ｉ）でのずれの大きさに相応する
ΔＲ＝｜Ｒ₁ −（Ｇ₁ ＋Ｈ₁ Ｌ_I Ｌ_O Ｃ₁ ）｜² 、又は、
ΔＲ＝｜Ｒ₂ −（Ｇ₁ ＋Ｈ₂ Ｌ_I Ｌ_O Ｃ₁ ）｜² と、
ΔＨ＝｜Ｈ₂ −Ｈ₁ ｜²
の大きさを考慮した上で、ＧとＣを更新する。定性的にいって、ΔＲが大きければ古いＣ₁ とＧ₁ は信頼できない。ΔＨが大きけれぼ新しいＧとＣは信頼できる。
【００８８】
そこで、ΔＲとΔＨの多寡を峻別するために予め定めた定数をαとβとし、ΔＲとΔＨの両方を考慮したパラメータとしてｒ＝ｍｉｎ（（αΔＲ＋βΔＨ），１）を求め、
Ｇ₂ ＝（１−ｒ）Ｇ₁ ＋ｒＧ (53)
Ｌ_I Ｌ_O Ｃ₂ ＝（１−ｒ）Ｌ_I Ｌ_O Ｃ₁ ＋ｒＬ_I Ｌ_O Ｃ (54)
で更新する。
§７−４．消音フィルタの再計算
（iv）Ｇ₂ とＣ₂ から消音フィルタ算出手順により消音フィルタの係数ｈ₃ を求める。ＦＦＴにより、ｈ₃ からＨ₃ を求めておく。この消音フィルタの係数ｈ₃ を使って消音を続行させる。すなわち、Ｌ_I Ｌ_O Ｃ₂ ，Ｇ₂ ，ｈ₃ ，Ｈ₃ をそれぞれＬ_I Ｌ_O Ｃ₁ ，Ｇ₁ ，ｈ₁ ，Ｈ₁ とおき直し、（ｉ）に戻る。
§８．消音対象周波数の帯域制限
次に、周波数帯域の制限について説明する。消音フィルタには、Ｌ_I Ｌ_O を逆に補正するような特性が入っている。Ｌ_I Ｌ_O は本来消音しなくてもよい周波数帯域で落ちるような特性にしてある筈なので、Ｌ_I Ｌ_O を逆に補正するのは無理な努力を重ねることになる。
【００８９】
また、二次音源２のスピーカは周波数帯域が限られている。
いま、消音する周波数範囲を表した伝達関数をＦとする。Ｆの振幅は、消音したい範囲で１とし、位相は０とする。例えば、５０〜２０００Ｈｚの範囲を消音したいなら、４次のＢＰＦ（バンドパスフィルタ）の振幅特性を参考にして、Ｆの振幅を周波数ｆ［Ｈｚ］の関数として表すと、

とする。
【００９０】
これを、消音フィルタに算入するには、消音フィルタ算出手順の計算において、Ｇの代わりにＧＦを使ってＨを求めればよい。ＧはＦＦＴに由来する飛び飛びの周波数の値を有するので、式(55)でＦの値を求めるときはＧの周波数に対応した値を計算しなければならない。
§９、消音効果の算出手順
最後に、消音効果を算出する手順について説明しておく。騒音の大きさは式(34)のＰで表されるが、消音波を出していないときのＰ_OFF と、消音波を出しているときのＰ_ONの比で、消音効果を利用者に示すことにする。
【００９１】
さらに、音響の分野ではｄＢ（デジベル）表示が一般的な点に配慮すると、消音効果Ｐ_ON/OFFは、
Ｐ_ON/OFF＝１０log₁₀ （Ｐ_ON／Ｐ_OFF ） (56)
となる。消音波を出さないときはＨ＝０と置けばよいことに着目し、式(35)を使って式(56)を展開すると、

で算出できる。
【００９２】
§４−２で前述したように＜｜Ｕ｜² ＞をそのまま＜｜Ｘ｜² ＞の代わりに使う場合は、式(33)の右辺の（１−γ² ）＜｜Ｄ｜² ＞をそのまま＜｜Ｎ｜² ＞の代わりに使うのが妥当である。
また、ナイキスト周波数に近くなるに従ってローパスフィルタ５の減衰が徐々に大きくなるのを補正する意味から、消音対象とする周波数においては１／｜Ｌ_I ｜² を＜｜Ｕ｜² ＞に乗じたものを＜｜Ｘ｜² ＞の代わりに使う場合には、１／｜Ｌ_I ｜² を（１−γ² ）＜｜Ｄ｜² ＞に乗じたものを＜｜Ｎ｜² ＞の代わりに使うのが妥当である。
【００９３】
なお、騒音のパワースペクトルの分布が平坦であることが判っているなら、＜｜Ｘ｜²＞は周波数に依存しない値となるので、式(56)において＜｜Ｘ｜²＞を乗ずる処理を省いてもよい。
また、周波数の重み付けＷが平坦、すなわち、周波数によらずに常に１ならば、式(57)においてＷを乗ずる処理を省けるのはいうまでもない。
【００９４】
＜発明の実施＞
図４は、本発明による能動騒音制御を実施する装置の構成を示す。
２１は騒音源２０からの騒音波を検出する参照マイクであり、２２は消音波を放出する二次音源となるスピーカである。騒音波と消音波が打ち消しあった後に残る残留波は、２３のエラーマイクで検出される。
参照マイク２１からの信号は、マイクアンプ２４、アンチエイリアシングのためのローパスフィルタ２５、ＡＤ変換器２６を経て信号ｘとなり、ＤＳＰ（ディジタル信号処理プロセッサ）２７に供給される。
【００９５】
ＤＳＰ２７から出力された信号ｙは、ＤＡ変換器２９、スムージングのためのローパスフィルタ３０、パワーアンプ３１を経て二次音源スピーカ２２に供給される。
エラーマイク２３からの信号は、マイクアンプ３２、アンチエイリアシングのためのローパスフィルタ３３、ＡＤ変換器３４を経て信号ｅとなり、ＤＳＰ２７に供給される。
【００９６】
サンプリングクロック供給回路２８はＡＤ変換器２６，３４、ＤＡ変換器２９にサンプリングクロックを供給するとともに、ＤＳＰ２７の割り込み端子にもこのサンプリングクロックを与える。
ＤＳＰ２７においては、ＡＤ変換器２６，３４、ＤＡ変換器２９の変換と同期して割り込みルーチンが起動される。
【００９７】
図３で示した消音フィルタ７は、図４においてはＤＳＰ２７の割り込みルーチンで実施され、図３で示した伝達関数算出部８と適応制御部１５は、図４においてはＤＳＰ２７のメインルーチンで実施される。
なお、図４においては説明の単純化のためローパスフィルタ等に起因するＬ_I とＬ_O による影響は、騒音波伝搬系Ｇと消音波伝搬系Ｃに含まれるとみなす。以降では、ＡＤ変換器２６を経た信号ｘを騒音波、ＤＡ変換器２９に供給する信号ｙを消音波、ＡＤ変換器３４を経た信号ｅを残留波と呼ぶことにする。すると、ｘからｅまでの経路が騒音波伝搬系Ｇであり、ｙからｅまでの経路が消音波伝搬系Ｃとなる。
【００９８】
なお、能動騒音制御の定式化の§１で前述した通り、定式化で示した式は、Ｌ_I ＝Ｌ_O ＝１と置けば、図４の実施例に帰着する。
また、二次音源から参照マイクヘの消音波の伝搬（フィードバック）がある場合（図５(a) 参照）には、この伝搬系の伝達関数をＢとおくと、図５(b) に示すように消音波伝搬系の特性をＣ−ＧＢとおくことによってフィードバックがない場合に帰着できる。すなわち、本発明は、フィードバックの有無にかかわらず適用が可能である。
【００９９】
ＤＳＰ２７内で行うメインルーチンと割リ込みルーチンの機能分担を模式的に図６に示す。
メインルーチン７１では、伝達関数算出と消音フィルタ係数算出を行う。割り込みルーチン７２では、ＡＤ変換器２６，３４からの騒音波と残留波の読み込み、消音波を作成する計算、ＤＡ変換器２９ヘの消音波の書き出しを行う。
【０１００】
メインルーチン７１と割り込みルーチン７２の間でデータをやりとりするために、データ受け渡し領域Ａ３，領域Ａ５，領域Ａ７がある。具体的には、メインルーチン７１は、新しい消音フィルタの係数を領域Ａ３に書き込む。割り込みルーチン７２は、領域Ａ３に新しいデータが揃ったら一括して、割り込みルーチン７２専用に確保してある領域Ａ４に書き移す。
【０１０１】
割り込みルーチン７２における消音波の計算には領域Ａ４の領域を使用するので、メインルーチン７１が領域Ａ３にデータを逐次書き込んでいる最中であっても、消音波の計算に支障はない。
騒音波と残留波のデータは割り込みルーチン７２から領域Ａ５と領域Ａ７に書き込む。騒音波と残留波のデータはクロススペクトル法による伝達関数の計算に使われるが、そのために必要な数のデータが領域Ａ５と領域Ａ７に集まると、メインルーチン７１は領域Ａ６と領域Ａ８に書き移す。以降は、メインルーチン７１は領域Ａ６と領域Ａ８を使って伝達関数の計算を進める。
【０１０２】
メインルーチン７１における伝達関数の計算は領域Ａ６と領域Ａ８を使って進むので、並行して割り込みルーチン７２が領域Ａ５と領域Ａ７に書き込んでも、伝達関数の計算には支障がない。
図７は、ＤＳＰ２７の割り込み処理のフローチャートである。
Ｓ１０とＳ１１は、新しい消音フィルタの係数が、メインルーチンによって領域Ａ３に用意されている場合に、それを領域Ａ４に移す処理である。
【０１０３】
Ｓ１２は、騒音波ｘ_n と残留波ｅ_n をそれぞれＡＤ変換器２６とＡＤ変換器３４から読み出す処理である。ここで、添字ｎは現在のサンプリング時刻におけるデータを意味し、ｎ−ｉはｉサンプリング周期前の時刻におけるデータを意味する。
Ｓ１４は、消音波ｙ_n を計算する処理である。消音フィルタの係数は全部でｐ個あるものとしている。
【０１０４】
Ｓ１５は消音波ｙ_n をＤＡ変換器２９に書き出す処理である。
図８と図９は、ＤＳＰ２７のメインルーチンにおける立ち上げ処理を示している。
図８のＳ２０で全作業領域の初期化を行う。
Ｓ２１は消音フィルタの係数としてｈ₁ を設定する処理である。
【０１０５】
Ｓ２２で割り込み許可を行っており、以降は、サンプリング周期ごとに図７の割り込み処理が起動されることになる。
Ｓ２３は、クロススペクトル法により伝達関数等を算出する処理である。
Ｓ２４とＳ２５は、消音フィルタの係数をｈ₂ に代えて、伝達関数等を算出する処理である。実際の消音動作に入るまでの期間は、消音波を出さないようにするためＳ２６では、消音フィルタの係数に０を設定している。
【０１０６】
Ｓ２７で騒音波伝搬系の伝達関数Ｇ₁ と消音波伝搬系の伝達関数Ｃ₁ を算出し、Ｓ２８で消音フィルタの係数ｈ₃ を算出している。
図９のＳ２９とＳ３０で消音量を算出し、Ｓ３１で利用者に表示する。このとき、消音量以外に、｜Ｘ₂ ｜² ，Ｇ₁ ，Ｃ₁ ，Ｈ₃ ，ｈ₃ ，γ₂ ² ，ｇ₁ （Ｇ₁ にＩＦＦＴを施すと得られる），ｃ₁ （Ｃ₁ にＩＦＦＴを施すと得られる）のいずれか又は全部を表示してもよい。
【０１０７】
利用者はＳ３２で、消音動作を行うか、断念するかを入力する。もし、消音動作を行う場合はＳ３３からＳ３４に進み、以降は図１２の消音の維持に進む。
なお、Ｓ３２で利用者に消音動作の可否を尋ねるのではなく、予め定めた閾値よりもよい消音効果が得られるならば、自動的にＳ３４に進める実施も可能である。
【０１０８】
図１０は、図８のＳ２３とＳ２５、図１２のＳ５７とＳ６４から呼び出されるクロススペクトル法による伝達関数等を算出するサブルーチンを示している。
Ｓ４０で、平均のために総和を求める作業領域等を初期化する。
Ｓ４１で割り込みルーチンからデータを受け取る領域Ａ５と領域Ａ７を初期化し、Ｓ４２でデータがＦＦＴに必要な数だけ集まるまで待ち、Ｓ４３でデータをメインルーチン専用の作業領域Ａ６と領域Ａ８に移す。
【０１０９】
Ｓ４４でハニング窓をかけ、Ｓ４５でＦＦＴによりスペクトルを算出する。
Ｓ４６はパワースペクトル、クロススペクトルの総和を求める処理であり、Ｓ４７とＳ４８で平均化回数ｋ₀ だけ繰り返しを行い、平均化回数ｋ₀ になればＳ４９の除算で平均を算出している。
その後Ｓ５０で、伝達関数Ｒとコヒーレンス関数γ² を算出する。
【０１１０】
なお、平均化回数ｋ₀は図８のＳ２０においてｋ₁に設定され、図１３のＳ６９と図１２のＳ６０ｃにおいてそれぞれｋ₂とｋ₁に変更される。ｋ₂＜ｋ₁なる大小関係にしておけば、平均化回数ｋ₀がｋ₁のときは測定時間が長いが精度がよくなり、平均化回数ｋ₀がｋ₂のときは精度が悪くなるが測定時間が短くて済む。
【０１１１】
図１１は、図８のＳ２８と、図１３のＳ６６と、図１４(a) のＳ７２と、図１４(b) のＳ７４から呼び出される、消音フィルタの計算を行うサブルーチンを示している。
Ｓ５１で自己相関関数ｒ（ｒ＝｛ｒ_0,ｒ_1,ｒ_2,‥‥ｒ_k-1 ｝とする）、相互相関関数ｄ（ｄ＝｛ｄ_0,ｄ_1,ｄ_2,‥‥ｄ_k-1 ｝とする）を計算する。この際に、重み付け係数Ｗなども算入しておく。
【０１１２】
Ｓ５２以降は、Levinsonアルゴリズムによる消音フィルタの係数ｈの算出を示している（消音フィルタの段数はｐとする）。
なお、Levinsonアルゴリズムについては、例えば、谷萩隆嗣、「ディジタル信号処理の理論 １基礎・システム・制御」、コロナ社、p.４４（1985）などで解説されているので、ここでは説明を省く。
【０１１３】
図１２及び図１３は、ＤＳＰ２７のメインルーチンにおいて消音を維持する処理を示している。
Ｓ５５は消音フィルタの係数ｈ₁ を領域Ａ３に設定し、この係数ｈ₁ による消音動作を開始するための処理である。後の処理のために、Ｓ５６では係数ｈ₁ を周波数領域で表したＨ₁ に変換している。
【０１１４】
騒音波伝搬系の伝達関数Ｇ₁ と消音波伝搬系の伝達関数Ｃ₁ とＨ₁ によれば、伝達関数はＧ₁ ＋Ｈ₁ Ｃ₁ になる筈だが、これと、Ｓ５７で求めた伝達関数Ｒ₁ とのずれｑをＳ５８で求め、閾値δとの比較をＳ５９で行っている。もし、ずれｑが閾値δ以下で、かつ、Ｓ６０ｃの判定においてクロススペクトル法の平均化回数ｋ₀ がｋ₁ ならば、消音状態は良好に保たれているので、このままの状態を維持することにし、消費電流の節約を狙ったＳ６０ｂに進む。
【０１１５】
Ｓ６０ｂでは割り込み処理を行っている以外の間はＤＳＰ２７をパワーダウンモードにする。
Ｓ６０ｂを抜けるのは、消音状態のずれが小さいままに留まると期待できる一定時間であり、音響系の構成にもよるが、例えば、１０秒などとする。
Ｓ６０ｃで平均化回数ｋ₀ がｋ₁ でない（すなわち、ｋ₂ である）ならば、Ｓ６０ｃで平均化回数ｋ₀ をｋ₁ に増やす（ｋ₂ ＜ｋ₁ ）ことでクロススペクトル法による測定の精度を高くしてから、Ｓ５７による測定を再度行うこととしている。
【０１１６】
Ｓ５９でずれｑが閾値δより大きい場合は、騒音波伝搬系の伝達関数Ｇ₁ と消音波伝搬系の伝達関数Ｃ₁ の更新が必要なので、Ｓ６１に進む。
Ｓ６１では決定した係数ｈ₁ とは異なる消音フィルタの係数ｈ₂ をＳ６２で設定し、これに対応する伝達関数をＳ６４で求める。
Ｓ６５でＧ₁ とＣ₁ を更新し、Ｓ６６で次回に使用する消音フィルタの係数ｈ₃ を算出する。
【０１１７】
Ｓ６７では、Ｓ５５からＳ６６までの処理を繰り返せるように、ｈ₃，Ｇ₂，Ｃ₂ をｈ₁，Ｇ₁，Ｃ₁に代入している。
Ｓ６８（とＳ６９）を経て、Ｓ５５ヘと戻ることにより、消音状態を維持する処理を繰り返し行う。
Ｓ６８とＳ６９では、ずれｑが閾値δより大きい場合はＳ６９で平均化回数ｋ₀をｋ₂に減らす（ｋ₂＜ｋ₁）ことで、クロススペクトル法によるＳ５７とＳ６４を再度行うときの測定時間を短くする。これは、消音状態を維持する繰り返し時間を短縮して、系の変動ヘの追従性を高めるためである。
【０１１８】
図１４には、図１２のＳ６１で行うベき消音フィルタの係数ｈ₂ の決定処理を３通り示してある。騒音波伝搬系の伝達関数Ｇと消音波伝搬系の伝達関数Ｃのどちらがずれやすいかは音響系の構成に依存する点等を勘案して、これらの中からどれかを実施すればよい。
図１４(a) は、騒音波伝搬系の伝達関数Ｇ₁ がＧにずれたと想定する場合に、Ｓ７２で係数ｈ₂ を算出する処理（§７−２（ii）▲１▼）を示している。
【０１１９】
図１４(b) は、消音波伝搬系の伝達関数Ｃ₁ がＣにずれたものと想定する場合に、Ｓ７４で係数ｈ₂ を算出する処理（§７−２（ii）▲２▼）を示している。
図１４(c) は、騒音波伝搬系の伝達関数Ｇ₁ がずれた場合を想定しているが、係数ｈ₁ に一つのインパルス波形だけを追加するという条件の元で適切な係数ｈ₂ を求める処理（§７−２（ii）▲３▼）を示している。
【０１２０】
図１５は、図１２のＳ６５から呼び出される、Ｇ₂ とＣ₂ の第１の計算処理（§７−３(iii) ▲１▼）を示している。Ｓ８１はＧ₁ とＣ₁ 等を構成する各複素数ごとに処理を行うためのループである。Ｓ８１では、
Ｈ₁ ＝｛Ｈ₁ ⁽⁰⁾ ，Ｈ₁ ⁽¹⁾ ，‥‥，Ｈ₁ ^(k-1) ｝
Ｈ₂ ＝｛Ｈ₂ ⁽⁰⁾ ，Ｈ₂ ⁽¹⁾ ，‥‥，Ｈ₂ ^(k-1) ｝
Ｒ₁ ＝｛Ｒ₁ ⁽⁰⁾ ，Ｒ₁ ⁽¹⁾ ，‥‥，Ｒ₁ ^(k-1) ｝
Ｒ₂ ＝｛Ｒ₂ ⁽⁰⁾ ，Ｒ₂ ⁽¹⁾ ，‥‥，Ｒ₂ ^(k-1) ｝
Ｇ₁ ＝｛Ｇ₁ ⁽⁰⁾ ，Ｇ₁ ⁽¹⁾ ，‥‥，Ｇ₁ ^(k-1) ｝
Ｃ₁ ＝｛Ｃ₁ ⁽⁰⁾ ，Ｃ₁ ⁽¹⁾ ，‥‥，Ｃ₁ ^(k-1) ｝
Ｇ₂ ＝｛Ｇ₂ ⁽⁰⁾ ，Ｇ₂ ⁽¹⁾ ，‥‥，Ｇ₂ ^(k-1) ｝
Ｃ₂ ＝｛Ｃ₂ ⁽⁰⁾ ，Ｃ₂ ⁽¹⁾ ，‥‥，Ｃ₂ ^(k-1) ｝
と表すものとする。Ｓ８５では、式(51)と式(52)を実施しており、Ｓ８６で式(47)と式(48)を実施している。
【０１２１】
なお、図１４で(c) を採用した場合は、ΔＨは添字ｉとは無関係の値になるので、Ｓ８２はＳ８１のループの前に出すことができる。
図１６は、図１２のＳ６５から呼び出される、Ｇ₂ とＣ₂ の第２の計算処理（§７−３(iii) ▲２▼）を示している。Ｓ９４におけるＨ₂ −Ｈ₁ の除算において分母が０になると計算に困る。そこで、図１２のＳ６１においては図１４(c) の処理を実施し、また、図１２のＳ５９によってずれがある場合に限ってＳ６１に進むようにしておくとよい。
【０１２２】
なお、Ｓ９１では、
ΔＲ＝｜Ｒ₁ −（Ｇ₁ ＋Ｈ₁ Ｃ₁ ）｜²
の代わりに
ΔＲ＝｜Ｒ₂ −（Ｇ₁ ＋Ｈ₂ Ｃ₁ ）｜²
を計算してもよい。
【０１２３】
＜請求項記載の発明との対応＞
最後に、請求項記載の発明の構成と、フローチャートの対応を示しておく。
(1) 請求項１記載の発明
消音フィルタに相異なる係数ｈ₁ と係数ｈ₂ を設定する手段は、図１２のＳ５５，Ｓ６１，Ｓ６２で実施されている。
係数ｈ₁ と係数ｈ₂ それぞれの場合に対応するＲ₁ とＲ₂ を測定する手段は、図１２のＳ５７とＳ６４で実施されている。
【０１２４】
Ｈ₁とＲ₁、Ｈ₂とＲ₂に関する連立方程式を解くことでＧとＣを算出する手段は、図１２のＳ６５で実施されている。
消音効果が最もよくなるように最小自乗規範に従って新たな消音フィルタの係数ｈ₃をＧとＣを使って算出する手段は、図１３のＳ６６で実施されている。
【０１２５】
消音動作中に相異なる消音フィルタの係数を設定しては、参照センサの信号からエラーマイクの信号ヘの伝達関数を測定し、騒音波の伝搬経路の伝達関数と消音波の伝搬経路の伝達関数を算出し、新たな消音フィルタの係数を算出し、この消音フィルタの係数を使って消音動作を行うこと、を繰り返すことは、図１２のＳ５５から図１３のＳ６７を回るループで実施されている。
【０１２６】
(2) 請求項２記載の発明
ＧとＣと＜｜Ｘ｜² ＞を算出する手段は、図１２のＳ５５〜Ｓ５７とＳ６１〜Ｓ６５で実施されている。
消音効果が最もよくなるように最小自乗規範に従って新たな消音フィルタの係数ｈをＧとＣを使って算出する手段であって、最小自乗規範に＜｜Ｘ｜² ＞の重み付けを有する手段は図１１で実施されており。特に＜｜Ｘ｜² ＞の重み付けはＳ５１における乗算で実施されている。
【０１２７】
算出された新たな消音フィルタの係数ｈを使って消音動作を行うことは、図１２、図１３においてＳ６６、Ｓ６７の後でＳ５５で消音フィルタの係数を設定することで実施されている。
【０１２８】
(3) 請求項３記載の発明
ＧとＣを算出する手段は、図１２のＳ５５〜Ｓ５７とＳ６１〜Ｓ６５で実施されている。
消音効果が最もよくなるように最小自乗規範に従って新たな消音フィルタの係数ｈをＧとＣを使って算出する手段であって、最小自乗規範にＷの重み付けを有する手段は図１１で実施されており、特にＷの重み付けはＳ５１における乗算で実施されている。
算出された新たな消音フィルタの係数ｈを使って消音動作を行うことは、図１２、図１３においてＳ６６，Ｓ６７の後でＳ５５で消音フィルタの係数を設定することで実施されている
(4) 請求項４記載の発明
ＧとＣを算出する手段は、図１２のＳ５５〜Ｓ５７とＳ６１〜Ｓ６５で実施されている。
【０１２９】
ＧにＦの重み付けを加えながら、消音効果が最もよくなるように最小自乗規範に従って新たな消音フィルタの係数ｈをＧとＣを使って算出する手段は、図１１で実施されており、特にＦの重み付けはＳ５１における乗算で実施されている。
算出された新たな消音フィルタの係数ｈを使って消音動作を行うことは、図１２、図１３においてＳ６６、Ｓ６７の後でＳ５５で消音フィルタの係数を設定することで実施されている。
【０１３０】
(5) 請求項５記載の発明
消音フィルタの係数ｈと記憶してあるＧとＣの値から予想されるＲの値（Ｒ₀ とする）を算出する手段は、図１２のＳ５８でＧ₁ ＋Ｈ₁ Ｃを計算することにより実施されている。
係数ｈを消音フィルタに設定して消音動作を行っているときのＲの値（Ｒ₁ とする）を測定する手段は、図１２のＳ５７で実施されている。
Ｒ₁ とＲ₀ のずれを予め定めた閾値と比較する手段は、図１２のＳ５９で実施されている。
ＧとＣを算出する手段は、図１２のＳ５５〜Ｓ５７とＳ６１〜Ｓ６５で実施されている。
【０１３１】
ＧとＣを使って新たな消音フィルタの係数ｈ₃ を算出する手段は、図１３のＳ６６で実施されている。
Ｒ₁ とＲ₀ のずれが予め定めた閾値よりも大きい場合に限ってＧとＣを算出し直し、この算出しなおしたＧとＣを使って消音フィルタの係数ｈ₃ を求め、この新たな消音フィルタの係数ｈ₃ を使って消音動作を行うことは、Ｓ５９でＮＯに分岐した後にＳ６１〜Ｓ６７からＳ５５に至る経路で実施されている。
【０１３２】
(6) 請求項６記載の発明
消音フィルタの係数ｈ₁ において消音動作を行っているときのＲの値（Ｒ₁ とする）を測定する手段は、図１２のＳ５７で実施されている。
記憶してあるＣ₁ とＨ₁ とＲ₁ から、参照センサからエラーマイクヘの騒音波の伝搬経路の伝達関数Ｇを算出する手段は、図１４(a) のＳ７１で実施されている。
【０１３３】
ＧとＣ₁ から新たな消音フィルタの係数ｈ₂ を算出する手段は、図１４(a) のＳ７２で実施されている。
算出された新たな消音フィルタの係数ｈ₂ を使って消音動作を行うことは、図１２のＳ６２で実施されている。
【０１３４】
(7) 請求項７記載の発明
消音フィルタの係数ｈ₁ において消音動作を行っているときのＲの値（Ｒ₁ とする）を測定する手段は、図１２のＳ５７で実施されている。
記憶してあるＧ₁ とＨ₁ とＲ₁ から、二次音源からエラーマイクヘの消音波の伝搬経路の伝達関数Ｃを算出する手段は、図１４(b) のＳ７３で実施されている。
Ｇ₁ とＣから新たな消音フィルタの係数ｈ₂ を算出する手段は、図１４(b) のＳ７４で実施されている。
【０１３５】
算出された新たな消音フィルタの係数ｈ₂ を使って消音動作を行うことは、図１２のＳ６２で実施されている。
【０１３６】
(8) 請求項８記載の発明
消音フィルタの係数ｈ₁ において消音動作を行っているときのＲの値（Ｒ₁ とする）を測定する手段は、図１２のＳ５７で実施されている。
記憶してあるＣ₁ とｈ₁ とＲ₁ から、ｈ₂ を算出する手段であって、消音効果がよりよくなるようにｈ₁ にインパルス波形を追加することでｈ₂ を算出する手段は、図１４(c) のＳ７５からＳ７８で実施されている。
算出された新たな消音フィルタの係数ｈ₂ を使って消音動作を行うことは、図１２のＳ６２で実施されている。
【０１３７】
(9) 請求項９記載の発明
消音フィルタに相異なる係数ｈ₁ と係数ｈ₂ を設定する手段は、図１２のＳ５５とＳ６１とＳ６２で実施されている。
係数ｈ₁ と係数ｈ₂ それぞれの場合に対応するＲ₁ とＲ₂ を測定する手段は、図１２のＳ５７とＳ６４で実施されている。
【０１３８】
消音フィルタの係数ｈ₁ ，Ｇ₁ ，Ｃ₁ から予想されるＲの値とＲ₁ のずれであるΔＲ、又は、消音フィルタの係数ｈ₂ ，Ｇ₁ ，Ｃ₁ から予想されるＲの値とＲ₂ のずれであるΔＲを算出する手段は、図１６のＳ９１で実施される。
Ｈ₁ とＲ₁ 、Ｈ₂ とＲ₂ に関する連立方程式を解くことでＧとＣを算出する手段は、図１６のＳ９４で実施されている。
【０１３９】
算出されたＧ，Ｃと、以前に記憶してあるＧ₁ ，Ｃ₁ を使って、消音フィルタの係数算出に使うべきＧ₂ ，Ｃ₂ を算出する手段は、図１６のＳ９５で実施されている。
消音効果が最もよくなるように最小自乗規範に従って新たな消音フィルタの係数ｈ₃ をＧ₂ とＣ₂ を使って算出する手段は、図１３のＳ６６で実施されている。
【０１４０】
ΔＲが予め設定してある値に比べて小さいときは、算出されたＧ，Ｃと、以前に記憶してあるＧ₁ ，Ｃ₁ をΔＲの大ききに関連させて内分することにより、Ｇ₂ ，Ｃ₂ を算出するという特徴は、図１６のＳ９３とＳ９５で実施されている。
Ｇ₂ ，Ｃ₂ を使って算出された新たな消音フィルタの係数ｈ₃ を使って消音動作を行うことは、図１３のＳ６６で実施されている。
【０１４１】
(10)請求項１０記載の発明
消音フィルタに相異なる係数ｈ₁ と係数ｈ₂ を設定する手段は、図１２のＳ５５とＳ６１とＳ６２で実施されている。
係数ｈ₁ と係数ｈ₂ それぞれの場合に対応するＲ₁ とＲ₂ を測定する手段は、図１２のＳ５７とＳ６４で実施されている。
【０１４２】
Ｈ₁とＲ₁、Ｈ₂とＲ₂に関する連立方程式を解くことでＧとＣを算出する手段は、図１２のＳ６５、すなわち図１５における、Ｓ８３とＳ８５の組み合せ、又は、Ｓ８３とＳ８６の組み合せで実施されている。
算出されたＧ，Ｃと、以前に記憶してあるＧ₁，Ｃ₁を使って、消音フィルタの係数算出に使うべきＧ₂，Ｃ₂を算出する手段は、図１２のＳ６５すなわち図１５で実施されている。
消音効果が最もよくなるように最小自乗規範に従って新たな消音フィルタの係数ｈ₃をＧ₂とＣ₂を使って算出する手段は、図１３のＳ６６で実施されている。
【０１４３】
ΔＨ＝｜Ｈ₂ −Ｈ₁ ｜² が予め設定してある値に比べて小さいときは、算出されたＧ，Ｃと、以前に記憶してあるＧ₁ ，Ｃ₁ をΔＨの大きさで内分することにより、Ｇ₂ ，Ｃ₂ を算出するという特徴は、図１５のＳ８４による判断の後Ｓ８６により実施されている。
ΔＨ＝｜Ｈ₂ −Ｈ₁ ｜² が予め設定してある値に比べて大きいときは、算出されたＧ，Ｃを、Ｇ₂ ，Ｃ₂ に用いるという特徴は、図１５のＳ８４による判断の後Ｓ８５により実施されている。
【０１４４】
Ｇ₂ ，Ｃ₂ を使って算出された新たな消音フィルタの係数ｈ₃ を使って消音動作を行うことは、図１３のＳ６６で実施されている。
算出された新たな消音フィルタの係数ｈを使って消音動作を行うことは、図１２、図１３においてＳ６６、Ｓ６７の後でＳ５５で消音フィルタの係数を設定することで実施されている。
【０１４５】
(11)請求項１１記載の発明
ＧとＣとγ² を算出する手段は、図８のＳ２１〜Ｓ２５とＳ２７で実施されている。
ＧとＣに基づいてＨを算出する手段は、図８のＳ２８と図９のＳ２９により実施されている。
【０１４６】
ＧとＣとγ² とＨを用いて消音量を算出する手段は、図９のＳ３０により実施されている。
算出された消音量を利用者に表示する手段は、図９のＳ３１により実施されている。
【０１４７】
(12)請求項１２記載の発明
参照センサの信号からエラーマイクへの信号ヘの伝達関数Ｒをクロススペクトル法によって求める手段であって、クロススペクトル法における平均化回数が可変である手段は、図１０のＳ４０からＳ５０までで実施されている。
消音のずれの程度を算出する手段は、図１２のＳ５８で実施されている。
測定されたＲに基づいて消音フィルタの係数ｈを算出する手段は、図１３のＳ６６で実施されている。
消音のずれが少ないときはより多い平均化回数でクロススペクトル法を行うことでＲを測定することは、図１２のＳ５９を経てＳ６０ｃで平均化回数をｋ₁ に設定した後、Ｓ５７とＳ６４が行われることで実施される。
【０１４８】
消音のずれが多いときはより少ない平均化回数でクロススペクトル法を行うことでＲを測定することは、図１３のＳ６８からＳ６９に進んで平均化回数がｋ₂ に設定された後、Ｓ５７とＳ６４が行われることで実施される。
測定されたＲに基づいて消音フィルタの係数ｈを算出し、この算出された係数ｈを使って消音動作をすることは、図１３のＳ６６のあとＳ６７を経てＳ５５が行われることで実施されている。
【０１４９】
【発明の効果】
(1) 請求項１記載の発明によれば、周波数領域で表された伝達関数から、漸近的な反復計算をせずに、所望のフィルタ長において最適な消音フィルタの係数が算出されるので、ひいては、系の変動への追従性能も高くなる。
所望のフィルタ長において最適な消音フィルタの係数を算出できるので、従来の能動騒音制御装置のように、単なる計算の都合で、フィルタ長を長くしたり、音響系を大型化する必要がなくなる。
【０１５０】
消音フィルタの係数の算出が短時間で行えるようになるため、消音動作中に、騒音伝搬系とキャンセル音伝搬系の伝達関数を算出し直しては消音フィルタの係数を再計算することが、短時間で繰り返して行える。従来の能動騒音制御装置では音響系はダクトなど固定的なものに限られていたが、本発明による能動騒音制御装置は音響系の変動に短時間で追従し消音を維持できる。そのため、エラーマイクを例えば人の頭部に配置しておいて、その人の動きに追従して消音させるというように新たな利用形態を実現することができる。
【０１５１】
(2) 請求項２記載の発明によれば、騒音波のスペクトル分布に配慮した消音を行える。そのため、全周波数範囲にわたって平坦とは限らない騒音波においては、よりパワーのある周波数を重点的に消音することができ、全体として消音効果が高くすることができる。
(3) 請求項３記載の発明によれば、周波数の重み付けをした消音を行える。そのため、全周波数範囲にわたって平坦な重み付けとは限らない場合、例えば、騒音が、Ａ特性の騒音計で評価される場合には、その周波数の重み付けに合わせたより適切な消音を行うことができる。
【０１５２】
(4) 請求項４記載の発明によれば、消音対象とする周波数範囲を制限できる。装置の構成上消音波を出せない周波数において無理に消音しなくて済むようになる。換言すれば、消音可能な周波数のみにおいて消音するよう制御されるので、無理や無駄が生じない。
(5) 請求項５記載の発明によれば、系の変動を検出できる。系の変動のない場合には不要な再測定を避けることができるので、省消費電力などに役立てることができる。
【０１５３】
(6) 請求項６記載の発明によれば、騒音波伝達関数が変動する場合に、消音効果を増やしつつ系を再測定する。従って、騒音波伝達関数が変動しやすい音響系においては、系を再測定する際に悪影響を与えることがない。
(7) 請求項７記載の発明によれば、消音波伝達関数が変動する場合に、消音効果を増やしつつ系を再測定する。従って、消音波伝達関数が変動しやすい音響系においては、系を再測定する際に悪影響を与えることがない。
【０１５４】
(8) 請求項８記載の発明によれば、後処理が容易な、系の再測定ができる。ｈ₂ −ｈ₁ がインパルス波形であり、周波数成分が全周波数帯域にわたるので、測定不能な周波数が存在しない。また、ｈ₂ −ｈ₁ がインパルス波形なので、ｈ₂ −ｈ₁ で除する演算を時間軸上で行うこともできるようになる。すなわち、連立方程式の解法の選択肢が多くなるので、装置を実現する際に都合のよい処理を選びやすくなる。
【０１５５】
(9) 請求項９記載の発明によれば、系の変動の程度に応じた系の算出ができる。系の変動が小さいときには以前に測定した結果を加味するので精度が高く、系の変動が大きいときは新たに測定した結果を重視するので追従性が高い。つまり、追従牲と高精度の兼ね合いを系の状況に合わせることができる。
(10)請求項１０記載の発明によれば、系の算出精度を勘案した系の算出ができる。新たに系を算出する精度の善し悪しに応じて、以前の算出結果を加味する。つまり、算出精度のバラツキによってある回の算出における精度の落ち込みがあらわになることを未然に防ぐことができる。
また、ＧとＣの算出の際に、Ｈ₂ −Ｈ₁ の大ききがほぼ０の場合において、Ｈ₂ −Ｈ₁ で除す演算を避けることができるので、計算に支障が生じることがない。
【０１５６】
(11)請求項１１記載の発明によれば、消音効果が定量的に表示される。すなわち、装置内に有する系の測定結果等を使って消音効果を表示するので、無理な音響系においては利用者は消音を断念することができ、無駄な努力をしなくて済む。また、利用者はよりより消音効果が得られるように音響系の構成を変えてみる（例えば、二次音源の位置を変える）ことができるようになる。
【０１５７】
(12)請求項１２記載の発明によれば、系の変動の程度に応じて系の測定精度が調節される。すなわち、系の変動が小さいときには測定に時間をかけて精度の高め、系の変動が大きいときは精度は劣るが測定の時間を滅らす。つまり、高精度と追従性の兼ね合いを系の状況に合わせることができる。
【図面の簡単な説明】
【図１】従来の一般的な能動騒音制御装置の構成を示す図である。
【図２】従来の、自動車の車室内の騒音低減を図る能動騒音制御装置の構成を示す図である。
【図３】本発明の能動騒音制御の定式化を説明するための能動騒音制御系の構成を示す図である。
【図４】本発明による能動騒音制御装置の構成を示す図である。
【図５】二次音源から参照マイクヘの消音波のフィードバックがある場合に、フィードバックがない系に帰着できることを説明する図である。
【図６】ＤＳＰ２７内で行われるメインルーチンと割リ込みルーチンの機能分担を模式的に示した図である。
【図７】ＤＳＰ２７の割り込み処理を説明するフローチャートである。
【図８】ＤＳＰ２７のメインルーチンにおける本発明の立ち上げ処理を説明するフローチャートである。
【図９】ＤＳＰ２７のメインルーチンにおける本発明の立ち上げ処理を説明するフローチャート（図８の続き）である。
【図１０】図８のＳ２３とＳ２５、図１２のＳ５７とＳ６４から呼び出される、クロススペクトル法による伝達関数等を算出する本発明のサブルーチンを示すフローチャートである。
【図１１】図８のＳ２８と、図１３のＳ６６と、図１４(a) のＳ７２と、図１４(b) のＳ７４から呼び出される、消音フィルタの計算を行うサブルーチンを示すフローチャートである。
【図１２】ＤＳＰ２７のメインルーチンにおいて消音を維持する本発明の処理を説明するフローチャートである。
【図１３】ＤＳＰ２７のメインルーチンにおいて消音を維持する本発明の処理を説明するフローチャート（図１２の続き）である。
【図１４】図１２のＳ６１で行うベき消音フィルタの係数ｈ₂ の本発明の決定処理を説明するフローチャートである。
【図１５】図１２のＳ６５から呼び出される、Ｇ₂ とＣ₂ の第１の計算処理（§７−３(iii) ▲１▼）を示すフローチャートである。
【図１６】図１２のＳ６５から呼び出される、Ｇ₂ とＣ₂ の第２の計算処理（§７−３(iii) ▲２▼）を示すフローチャートである。
【符号の説明】
１ 参照マイク
２ 二次音源
３ エラーマイク
４ マイクアンプ
５ ローパスフィルタ
６ ＡＤ変換器
７ 消音フィルタ
８ 伝達関数算出部
９ ＤＡ変換器
１０ ローパスフィルタ
１１ パワーアンプ
１２ マイクアンプ
１３ ローパスフィルタ
１４ ＡＤ変換器
１５ 適応制御部
２０ 騒音源
２１ 参照マイク
２２ スピーカ
２３ エラーマイク
２４ マイクアンプ
２５ ローパスフィルタ
２６ ＡＤ変換器
２７ ＤＳＰ（ディジタル信号処理プロセッサ）
２８ サンプリングクロック供給回路
２９ ＤＡ変換器
３０ ローパスフィルタ３０
３１ パワーアンプ３１
３２ マイクアンプ３２
３３ ローパスフィルタ
３４ ＡＤ変換器[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an active noise control apparatus that actively cancels a noise wave by canceling the noise wave with the sound cancellation, and in particular, the transfer characteristics of the noise wave propagation system and the silencer propagation system change during the silencing operation. However, the present invention relates to an active noise control device that keeps a good silenced state following the change.
[0002]
[Prior art]
The active noise control device is a device that reduces noise by generating a sound wave that cancels noise waves generated from a noise source.
According to a general configuration of a conventional active noise control apparatus, the active noise control apparatus includes a reference microphone that detects a noise wave generated from a noise source, a secondary sound source that emits a muffler, and a noise wave after mute. And an error microphone for detecting (residual wave).
[0003]
The silencer is created by passing a noise wave or a signal highly correlated with the noise wave through a silencer filter. The noise wave after mute (residual wave) is monitored by an error microphone, and the content of the mute filter is adaptively controlled so that the mute state is kept good.
The configuration of a conventional general active noise control device is shown in FIG. 1 (Toshifumi Kosaka “Practical technology of adaptive filter (5) – Active noise reduction using frequency domain adaptive algorithm”), Journal of the Acoustical Society of Japan, Vol. 48, 7 No., pp. 520-524 (1992)).
[0004]
In FIG. 1, 101 is a noise source. Reference numeral 102 is a reference microphone that detects noise waves, 103 is a secondary sound source (specifically, a speaker) that emits a silenced sound, and 104 is an error microphone that detects a noise wave (residual wave) after being silenced. The sound deadening is created by subjecting the noise wave picked up at 102 to signal processing by a FIR (Finite Impulse Response) digital filter 105. Reference numeral 106 denotes a controller that observes the noise wave detected by the reference microphone 102 and the residual wave detected by the error microphone 104, and updates the coefficient of the digital filter 105 so that the sound muting is performed more appropriately.
[0005]
This controller observes the D and M signals, and the transfer function R from D to M_MDThe transfer function T of the digital filter 105 such that is close to 0_FSearch for. More specifically, T at a certain point in time_FAnd R_MDBased on the value of T_FRepeat the correction of the value of.
As another conventional technique, an active noise control device for reducing noise in a passenger compartment of an automobile is also known (see JP-A-7-253790). The configuration of this active noise control device is shown in FIG.
[0006]
In FIG. 2, reference numeral 120 denotes a noise canceling system, which generates a noise canceling signal (dead sound) so that the power of an error signal (residual wave) is minimized. 123 is a noise canceling sound Sc_nThe canceling speaker (secondary sound source) that radiates the noise, 124 is arranged at the noise canceling point, and the noise S_nAnd canceling sound Sc_nThe synthesized sound is detected, and the synthesized sound signal is converted into the error signal e._nIt is an error microphone that outputs as
[0007]
121 is a reference signal x_nThe reference signal generator 122 generates a reference signal x generated from the reference signal generator 121._nIs input and the noise S at the noise cancellation point in the passenger compartment_nAnd canceling sound Sc_nIs the error signal e_nAnd the error signal e_nThe noise cancellation signal y is obtained by performing adaptive signal processing so that the_nIs output. The noise cancellation controller 122 includes an adaptive signal processing unit 122a, an adaptive filter 122b having a digital filter configuration, and a reference signal x._nIn addition, the propagation characteristics (transfer function) C of the canceling sound propagation system_MIs a signal processing reference signal (filtered reference signal) r_nAnd a signal processing filter 122c for generating
[0008]
  130Is a transfer function calculation processing unit for calculating simultaneous equations related to the transfer function of the canceling sound propagation system, and the reference signal x_nAnd error signal e_nAnd the coefficient w of the adaptive filter 122b_nThe transfer function c of the canceling sound propagation system using,The calculation is performed by iterative calculation that gradually reduces the error, and the signal processing filter 122c is sequentially set.
[0009]
SUMMARY OF THE INVENTION Problems to be Solved by the Invention and Means for Solving the Problems
Generally, in the active noise control device, the characteristics of the secondary sound source are superimposed when the sound deadening is emitted from the secondary sound source. Therefore, in order to create a silenced sound that cancels out with the noise wave, the silencer filter must have the inverse characteristics of the secondary sound source. The reverse characteristic of the secondary sound source becomes easier to realize as the delay time is allowed by separating the distance from the detection microphone to the secondary sound source. Or, if the delay time is not allowed much, the accuracy of the reverse characteristic must be given up.
[0010]
In the conventional general active noise control apparatus (FIG. 1) described above, it is assumed that the input M = 0 of the controller 106 is established at the time of mute, and the target of active mute is to set M = 0. However, the more M = 0 is established, the more accurate the reverse characteristic of the secondary sound source 103 needs to be realized, and the interval between the reference microphone 102 and the secondary sound source 103 must be increased. This leads to an increase in the size of the active noise control device.
[0011]
Further, the transfer function T of the digital filter 105 expressed in the frequency domain._FIf IFFT (Inverse Fast Fourier Transform) is calculated, in general, T_FThis impulse response should be the coefficient of the necessary FIR digital filter 105, but there is no guarantee that the length of this impulse response will fit within the length of the FIR digital filter 105 (also called the number of taps). . More specifically, for example, if the IFFT score is 1024, T_FImpulse response of 1024 points, FFT (Fast Fourier Transform) and IFFT are premised on periodic functions, so the first 512 points correspond to the positive time range, the second 512 points is the negative time range It is usual to make it correspond to. On the other hand, the length of the FIR digital filter 105 is limited by the capability of the signal processing system._FIt is convenient that the non-zero value in the impulse response is within the 300th range from the top. However, in the impulse response obtained by IFFT, in reality, it is normal that a value other than 0 appears after the 301st.
[0012]
However, if the filter length is long enough, the cost will increase, and in order to move the impulse response in the negative range to the positive range, the distance between the detection microphone and the secondary sound source must be increased. This will increase the size of the acoustic system.
As described above, the measures for calculating the inverse characteristic based on the transfer function expressed in the frequency domain have not been specifically formulated in the past. For example, Toshiro Oga, Yoshio Yamazaki, Yutaka Kaneda, "Sound System and Digital Processing" published by the Institute of Electronics, Information and Communication Engineers (1995), "Inverse calculation in the frequency domain causes errors if the filter length is not long enough. Careful consideration is necessary for the operation "(p. 156, lines 6-7).
[0013]
As described above, in the conventional general active noise control device (FIG. 1), the silencing filter T in the frequency domain._FIn order to promote this calculation, it is inevitable to increase the length of the muffler filter and increase the size of the acoustic system.
Further, unlike the active noise control device, the active noise control device (FIG. 2) for reducing the noise in the passenger compartment of the automobile described above performs all calculations in the time domain without converting to the frequency domain. For this reason, it is not necessary to lengthen the muffler filter length or to obtain a delay time. At the same time, however, it is difficult to solve the simultaneous equations and to calculate the canceling sound transfer function c, which requires calculation time in that it must be solved asymptotically by iterative calculation.
[0014]
In the present invention, a noise wave or a signal highly correlated with the noise is detected by a reference sensor, and the detected signal is passed through a silencing filter having a coefficient h to generate the silencing sound, and the silencing sound is emitted from the secondary sound source. Thus, in the active noise control device for reducing the residual wave at the position of the error microphone, it is possible to overcome these disadvantages in the conventional technology by revising the formulation of active noise control, It has the following objects and configurations.
[0015]
  (1) From the transfer function expressed in the frequency domain, it is possible to calculate the optimum mute filter coefficient for the desired filter length without performing asymptotic iterations. Increase performance.
  In order to achieve the above object, according to the first aspect of the present invention, the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is represented by G, and the transfer function of the propagation path of the silenced wave from the secondary sound source to the error microphone. Is represented by C, H represents the coefficient h of the mute filter as a transfer function, and R represents the measurement of the transfer function from the reference sensor signal to the error microphone signal.
  Different coefficients h for the silencer filter₁, And coefficient h₂And the coefficient h₁And coefficient h₂R corresponding to each case₁And R₂Means for measuring H, and H₁And R₁, H₂And R₂Means for calculating G and C by solving simultaneous equations and a coefficient h of a new silencing filter according to the least squares standard so that the silencing effect is the best_ThreeTheGWhenCIt is comprised so that it may have a means to calculate using (Claim 1).
[0016]
Set different coefficients of the muffler filter during the mute operation, measure the transfer function from the reference sensor signal to the error microphone signal, transfer function of noise wave propagation path and transfer function of mute wave propagation path Is calculated, the coefficient of the new silencing filter is calculated, and the silencing operation is repeated using the coefficient of the silencing filter. That is, the mute state is maintained as described in §7-2 to §7-4.
[0017]
Thereby, four arithmetic operations of the transfer function can be easily performed in the frequency domain. By solving simultaneous equations, the noise wave propagation system G and the silencer propagation system C can be calculated. The coefficient h of the silencing filter H is obtained from G and C by the least square criterion (the actual calculation is based on a general inverse matrix). The mute state can be maintained by repeating these frequently.
[0018]
(2) Perform noise reduction considering the spectrum distribution of noise waves.
In general, the spectrum distribution of a noise wave is not necessarily flat over the entire frequency range. If it thinks simply, although it is thought that the silencing effect will become high as a whole if the frequency which has a more powerful power is silenced, the spectrum distribution of such a noise wave is not considered in the conventional active noise control apparatus.
In order to achieve the above object, the invention according to claim 2 is characterized in that the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, and the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone. Is C and the spectrum distribution of the noise wave signal is <| X |²>
G and C and <| X |²> And means for calculating a new silence filter coefficient h using G and C according to the least-squares criterion so that the silencing effect is the best, and the least-squares criterion is <| X |²It is comprised so that it may have a means with a weight of> (Claim 2).
[0019]
According to the second aspect of the present invention, when calculating the silencing filter using G and C, <| X |²A mute operation is performed using a mute filter obtained by adding>. That is, <| X | described later in §5-2.²> Is included.
In the active noise control device according to the present invention, it is possible to mute in consideration of the deviation of the spectrum distribution of the noise wave, and thus a more appropriate muffling effect can be obtained.
[0020]
(3) Enable mute with frequency weighting.
Optimum mute does not necessarily have to be flat weighted over the entire frequency range. For example, noise is often evaluated with an A-weighted sound level meter. As in the case of the A characteristic, the noise reduction when the frequency weighting is taken into consideration is not taken into consideration in the active noise control device according to the prior art.
[0021]
In order to achieve the above object, the invention according to claim 3 is characterized in that the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, and the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone. Is C, and frequency weighting when evaluating the silencing effect is W,
Means for calculating G and C, and means for calculating a new mute filter coefficient h using G and C according to the least squares standard so that the silencing effect is the best, with the least squares standard weighted by W Means are provided (claim 3).
[0022]
According to the third aspect of the present invention, when calculating the silencing filter using G and C, the silencing operation is performed using the silencing filter obtained by adding W. That is, W is counted as described later in §5-2.
In the active noise control device according to the present invention, it is possible to mute in consideration of such frequency weighting, so that a more appropriate muffling effect can be obtained.
[0023]
(4) It is possible to limit the frequency range to be silenced.
A speaker as a secondary sound source does not have a reproduction capability in the entire frequency range, and it is difficult to output a silencer particularly at frequencies of several tens of Hz or less.
Further, due to the presence of the low-pass filters 110 and 112 (FIG. 1), there is an upper limit in the frequency band that can be silenced. It is useless to try to mute the sound while ignoring such frequency band limitations.
[0024]
In the active noise control device according to the present invention, a reasonable mute operation can be performed by taking into consideration the limitation of the frequency range.
In order to achieve the above object, the invention according to claim 4 is characterized in that the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, and the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone. Is C and F is frequency weighting indicating the frequency range to be muffled,
There is a means for calculating G and C, and a means for calculating a new silence filter coefficient h using G and C in accordance with the least-squares standard so that the silencing effect is best while adding weighting F to G. (Claim 4).
According to the fourth aspect of the present invention, when calculating the silencing filter using G and C, the silencing operation is performed using the silencing filter obtained by adding F to G. That is, the band limitation F described later in §8 is included.
[0025]
(5) Make system fluctuations detectable.
In order to achieve the above object, the invention according to claim 5 is characterized in that the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, and the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone. Is C, and R is a measurement of the transfer function from the reference sensor signal to the error microphone signal.
R value R expected from the coefficient h of the mute filter and the stored G and C values. And R value R when the coefficient h is set in the silencing filter and the silencing operation is performed.₁Means for measuring R, and R₁And R₀A means for comparing the deviation with a predetermined dark value, a means for calculating G and C, and a coefficient h of a new silence filter using G and C_ThreeAnd a means for calculating.
[0026]
In the invention of claim 5, R₁And R₀G and C are recalculated only when the deviation is larger than a predetermined threshold, and the coefficient h of the silence filter is calculated using the recalculated G and C._ThreeAnd the coefficient h of this new silence filter_ThreeUse the to mute the sound. That is, as described later in §7-1, “system deviation is detected and updated only when the deviation is large” is performed.
[0027]
In the active noise control device according to the present invention, the measurement result of the noise wave propagation system and the silencer propagation system is stored in the active noise device, and the fluctuation of the system can be detected by comparing the silenced state. become. As a result, unnecessary remeasurement can be avoided, which can be used for power saving and the like.
[0028]
(6) When the noise wave propagation system fluctuates, the system can be remeasured while increasing the silencing effect.
In order to achieve the above-mentioned object, the invention according to claim 6 is characterized in that the transfer function of the muffler propagation path from the secondary sound source to the error microphone is represented by C, and the coefficient h of the mute filter is represented by the transfer function as H, When R is the measurement of the transfer function from the reference sensor signal to the error microphone signal,
Silencer filter coefficient h₁Value of R when mute operation is performed in₁Means for measuring and C stored₁And H₁And R₁To calculate the transfer function G of the propagation path of the noise wave from the reference sensor to the error microphone, and G and C₁To a new silence filter coefficient h₂It is comprised so that it may have a means to calculate (Claim 6).
[0029]
In the invention of claim 6, R₁G is calculated on the basis of G and C₁Is used to calculate the coefficient of the mute filter and perform the mute operation. That is, based on §7-2 (ii) (1), the muffler filter is changed so as to follow the change of the system.
In the active noise control apparatus according to the present invention, in an acoustic system in which it is known in advance that the noise wave propagation system is likely to fluctuate, there is no adverse effect when the system is remeasured.
[0030]
(7) When the sound deadening propagation system fluctuates, re-measure the system while increasing the sound deadening effect.
In order to achieve the above object, the invention according to claim 7 is characterized in that the transfer function of the noise wave propagation path from the reference sensor to the error microphone is represented by G, and the coefficient h of the mute filter is represented by the transfer function as H. When R is the measurement of the transfer function from the sensor signal to the error microphone signal,
Silencer filter coefficient h₁Value of R when mute operation is performed in₁And means for measuring the stored G₁And H₁And R₁To calculate a transfer function C of the propagation path of the sound deadening from the secondary sound source to the error microphone;₁And a new silence filter coefficient h from C and C₂It is comprised so that it may have a means to calculate (Claim 7).
[0031]
In the invention of claim 7, R₁C is calculated based on₁And C are used to determine the coefficient of the mute filter, and the mute operation is performed. That is, based on §7-2 (ii) (2), the muffler filter is changed so as to follow the change of the system.
In an acoustic system in which it is known in advance that the silencer propagation system is likely to fluctuate, there is no adverse effect when the system is measured again.
[0032]
(8) Enables re-measurement of the system for easy post-processing.
In order to achieve the above object, the invention according to claim 8 is characterized in that the transfer function of the muffler propagation path from the secondary sound source to the error microphone is represented by C, and the coefficient h of the mute filter is represented by the transfer function as H, When R is the measurement of the transfer function from the reference sensor signal to the error microphone signal,
Silencer filter coefficient h₁Value of R when mute operation is performed in₁Means for measuring and C stored₁And coefficient h₁And R₁To the coefficient h₂Is a means for calculating the coefficient h so that the silencing effect is improved.₁By adding an impulse waveform to the coefficient h₂It is comprised so that it may have a means to calculate (Claim 8).
[0033]
In the invention of claim 8, C₁And coefficient h₁And R₁To the coefficient h₁Coefficient h with impulse waveform added to₂Is calculated and mute operation is performed. That is, based on §7-2 (ii) (3), the coefficient h₂And coefficient h₁The muffler filter is changed so that the difference between them becomes an impulse waveform.
In the active noise control device according to the present invention, the system can be re-measured so that the post-processing for calculating the system is easy while the silencing operation is performed.
[0034]
(9) The system can be calculated according to the degree of system fluctuation.
In order to achieve the above object, the invention according to claim 9 is characterized in that the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, and the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone. Is represented by C, H represents the coefficient h of the muffler filter as a transfer function, and R represents the measurement of the transfer function from the reference sensor signal to the error microphone signal.
Different coefficients h for the silencer filter₁And coefficient h₂And the coefficient h₁And coefficient h₂R corresponding to each case₁And R₂And the coefficient h of the muffler filter₁, G₁, C₁Expected R value and R₁Deviation or coefficient h of the silence filter₂, G₁, C₁Expected R value and R₂Means for calculating ΔR, which is a deviation of H, and H₁And R₁, H₂And R₂Means for calculating G and C by solving simultaneous equations with respect to G, C, and previously stored G₁, C₁G should be used to calculate the coefficient of the muffler filter₂, C₂And a new silence filter coefficient h according to the least squares standard so that the silencing effect is best._ThreeG₂And C₂It is comprised so that it may have a means to calculate using (Claim 9).
[0035]
In the invention of claim 9, when ΔR is smaller than a preset value, the calculated G and C and the previously stored G₁, C₁Is divided internally in relation to the magnitude of ΔR, G₂, C₂To calculate G₂, C₂New silence filter coefficient h calculated using_ThreeUse the to mute the sound. That is, G and C are updated as described later in §7-3 (iii) (2).
[0036]
When the system fluctuation is small, the accuracy is improved by taking into account the previously measured result. When the system fluctuation is large, the followability is improved by placing importance on the newly measured result. The active noise control apparatus according to the present invention having such a function can match the balance between the followability and the high accuracy to the situation of the system.
[0037]
(10) The system can be calculated in consideration of the calculation accuracy of the system.
In order to achieve the above object, the invention according to claim 10 is characterized in that the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, and the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone. Is represented by C, H represents the coefficient h of the mute filter as a transfer function, and R represents the measurement of the transfer function from the reference sensor signal to the error microphone signal.
Different coefficients h for the silencer filter₁And coefficient h₂And the coefficient h₁And coefficient h₂R corresponding to each case₁And R₂Means for measuring H, and H₁And R₁, H₂And R₂Means for calculating G and C by solving simultaneous equations with respect to G, C, and previously stored G₁, C₁G should be used to calculate the coefficient of the muffler filter₂, C₂And a new silence filter coefficient h according to the least squares standard so that the silencing effect is best._ThreeG₂And C₂It is comprised so that it may have a means to calculate using (Claim 10).
[0038]
In the invention of claim 10, ΔH = | H₂-H₁｜²Is smaller than the preset value, the calculated G and C and the previously stored G₁, C₁Is internally divided by the magnitude of ΔH, G₂, C₂And ΔH = | H₂-H₁｜²Is larger than a preset value, the calculated G and C are changed to G₂, C₂G₂, C₂New silence filter coefficient h calculated using_ThreeUse the to mute the sound. That is, G and C are updated as described later in §7-3 (iii) (1).
[0039]
The previous calculation results are taken into account according to the accuracy of calculating a new system. The active noise control apparatus according to the present invention having such a function can prevent a drop in accuracy in a certain calculation due to variations in calculation accuracy.
[0040]
(11) It is possible to display the silencing effect quantitatively.
In order to achieve the object, the invention according to claim 11 provides:
The transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone is C, the transfer function of the silence filter is H, and the noise wave in the error microphone The coherence function indicating the proportion of power derived from²And when
G, C and γ²Means for calculating H based on G and C, G, C and γ²And a means for calculating the muffled sound volume using H, and a means for displaying the calculated muffled sound volume to the user (claim 11).
[0041]
In the invention described in claim 11, the mute filter is calculated based on the result of measuring the system, and the mute volume is calculated and displayed to the user. That is, the muffled sound volume is calculated as described later in §9.
Since the active noise control device according to the present invention displays the silencing effect using the measurement results of the system in the device, the user can give up the silencing in an unreasonable acoustic system, making a wasteful effort. No need.
[0042]
(12) Adjust the measurement accuracy of the system according to the degree of system fluctuation.
In order to achieve the above object, the invention according to claim 12 is means for obtaining a transfer function R from a signal of a reference sensor to a signal to an error microphone by a cross spectrum method, wherein the number of times of averaging in the cross spectrum method is There are means that are variable, means for calculating the degree of muffled deviation, and means for calculating the coefficient h of the muffler filter based on the measured R (claim 12).
[0043]
In the invention of claim 12, when the deviation of the muffling is small, the R is measured by performing the cross spectrum method with a larger number of averaging, and when the deviation of the muffing is large, the cross spectrum method is performed with a smaller number of averaging. R is measured to calculate the coefficient h of the silencing filter based on the measured R, and the silencing operation is performed using the calculated coefficient h.
[0044]
In other words, by controlling the number of times of averaging in accordance with the amount of muffled deviation based on the relationship between the accuracy of the number of averaged times described later in §2 and the time required for observation, the accuracy is improved when the number of muffled deviations is small. If there is a lot of noise reduction, the time required for observation will be shortened.
When the system fluctuation is small, it takes time to improve the accuracy, and when the system fluctuation is large, the accuracy is inferior but the measurement time is reduced. The active noise control apparatus according to the present invention having such a function can match the balance between high accuracy and followability to the system conditions.
[0045]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
<Formulation of active noise control>
As a premise for deriving the embodiment of the present invention, the formulation of active noise control according to the present invention will be described from §1 to §9.
[0046]
§1. System configuration
An active noise control system to which the present invention is applied is shown in FIG.
Reference numeral 1 denotes a reference microphone for detecting a noise wave from a noise source (not shown). A signal detected by the reference microphone 1 is represented as x. Reference numeral 2 denotes a secondary sound source that emits a silenced sound. A signal given to the secondary sound source 2 is represented by y. The residual wave remaining after the noise wave and the sound cancellation are canceled by the three error microphones.
The residual wave also includes external noise n that is uncorrelated with noise. A signal detected by the error microphone 3 is represented by e.
[0047]
The noise wave propagation system when the noise wave is transmitted from the reference microphone 1 to the error microphone 3 is represented by G, and the sound wave propagation system when the sound wave is transmitted from the mute speaker to the error microphone 3 is represented by C.
Note that the noise wave propagation system G and the muffling propagation system C include characteristics of a microphone and a speaker. When a reference signal correlated with a noise wave is picked up by a sensor (for example, a vibration sensor) other than the microphone, a propagation system from the reference signal to the noise wave is also included in G.
[0048]
A signal x from the reference microphone 1 is output as a signal u through a microphone amplifier 4, a low-pass filter 5 for anti-aliasing, and an AD converter 6. This signal u is supplied to the mute filter 7 and the transfer function calculator 8.
The signal v output from the muffler filter 7 is output as a signal y through a DA converter 9, a low-pass filter 10 for smoothing, and a power amplifier 11, and is supplied to the secondary sound source 2.
[0049]
A signal e from the error microphone 3 is output as a signal d through a microphone amplifier 12, a low-pass filter 13 for anti-aliasing, and an AD converter 14, and is supplied to the transfer function calculation unit 8.
Based on the transfer function R measured by the transfer function calculation unit 8, the adaptive control unit 15 keeps the content of the silence filter 7 optimally.
[0050]
Representations in the frequency domain of x, y, e, u, v, d, and n are X, Y, E, U, V, D, and N, respectively. For example, U is obtained by performing FFT (Fast Fourier Transform) on u, and is a vector having a complex value at each frequency. As is well known, the number of powers of 2 sufficient to represent the system (for example, 512, 1024, 2048) is selected as the FFT score.
[0051]
  In addition, you may perform the process which applies a window function to u prior to FFT. Since the FFT is used to express the signal in the frequency domain, the resolution of the frequency when U is expressed is “sampling frequency” / “number of FFT points”, and U has a value for the skipped frequency. Needless to say.
  The transfer characteristics of the microphone amplifier 4, the low-pass filter 5, and the AD converter 6 and the transfer characteristics of the microphone amplifier 12, the low-pass filter 13, and the AD converter 14 are L._IThe transfer characteristic of the DA converter 9, the low-pass filter 10, and the power amplifier 11 is L_OAnd Then
  U = XL_I                                               (twenty one)
  V = UH (22)
  Y = VL_O                                               (twenty three)
  E = XG + YC+ N (24)
  D = EL_I                                               (twenty five)
Holds.
  For simplicity, L_I  , L_OCan be considered to be included in the noise wave propagation system G and the silencer propagation system C. In this case, u, v, and d may be referred to as noise wave, silencer, and residual wave, respectively._I= L_O= 1. In the configuration and embodiments of the present invention, description will be given below based on such simplification.
[0052]
§2. Calculation method of transfer function
Here, a transfer function calculation method in the transfer function calculation unit 8 will be described. There are various known methods for calculating the transfer function, but for the purpose of the following description, the cross spectrum method will be described as an example. Erasing X, V, E, Y, E from equations (21)-(25)
D = U (G + HL_IL_OC) + NL_I                  (26)
It becomes. Complex conjugate of U on both sides^*Multiply by U
    ^*UD =^*UU (G + HL_IL_OC) +^*UNL_I        (27)
It becomes. External noise N is^*Uncorrelated with U, so^*When the cross spectrum is obtained by averaging UD many times,^*UNL_IBecomes negligible compared to the first term on the right side. The process for obtaining this time average is represented by <>.
<^*UNL_I> = 0 (28)
Can be written.
[0053]
If the number of times for averaging (the number of times of averaging) is small, the accuracy deteriorates, but the observation time is shortened because only a small number of samples are required. Conversely, if the number of times of averaging is increased, more samples are required, so that the time required for observation becomes longer, but the accuracy is improved. For example, a value such as 20 or 100 is used as the averaging count.
Now, paying attention to the point that Equation (28) holds, taking the time average of both sides of Equation (27),
<^*UD> = <^*UU> (G + HL_IL_OC) (29)
It becomes. The transfer function calculation unit 8 converts the transfer function R into
R = <^*UD> / <^*UU> (30)
Is calculated by
R = G + HL_IL_OC (31)
It becomes. The transfer function calculator 8 is a coherence function γ.²Can also be calculated based on the following equation.
[0054]
γ²= <^*UD>²/ (<^*UU> <^*DD>) (32)
This coherence function gives the ratio of the power of the signal correlated with U included in the signal D, and takes a value from 0 to 1. 1-γ²Is the ratio of power due to external noise N.
<| NL_I｜²> = (1-γ²) <| D |²> (33)
Γ²When = 0, the power of the signal D is all due to the external noise N, and γ²When = 1, it means that the power of the signal D is correlated with the signal U.
[0055]
§3. Silence goal
§3-1. Evaluation of noise reduction by power
Next, the mute target will be described.
E representing the signal e in the frequency domain is a vector having a complex value for each frequency.
Now, | E |²Indicates that the power for each frequency of E is to be obtained (noting that E is a complex number, | E |²=^*EE)), <| E |²> Represents the time average, and Σ represents an operation for calculating the sum over all frequencies.
[0056]
The mute target is to minimize the square of the value for the signal e, that is, to use the least square criterion. Specifically, it will be described in §3-3.
[0057]
§3-2. Frequency weighting in the evaluation of noise reduction
When considering the quality of muffling, it is not always common to consider all frequencies with equal weighting. For example, a sound level meter is used to measure the degree of noise, and its A characteristic shows a weighting peak around 2500 Hz.
In addition, the C characteristic of the sound level meter is less weighted at 20 Hz or less or 12,500 Hz or more. Human auditory characteristics are also not flat frequency characteristics.
Therefore, the noise level is given a weight W for each frequency,
P = ΣW <| E |²> (34)
I will compare them. It is appropriate to select W according to the purpose of active silencing and the evaluation method. For example, if it is important to measure with the A characteristic of the sound level meter, W may be given the same weight as the A characteristic of the sound level meter.
[0058]
It should be noted that the expression expressing the weighting of the A characteristic of the sound level meter is, for example, as a function of the frequency f [Hz] according to the international standard IEC651, “Sound Level Meter Amendment 1” p. 15 (1993).
W (f) = 1200²f^Four(F²+20.6²)^-1(F²+12200²)^-1(F²+107.7²)^-1/2(F²+737.9²)^-1/2
(For the sake of simplicity, the normalization with a gain of 0 [dB] at 1000 [Hz] is omitted).
[0059]
Needless to say, <| E | in formula (34)²> Has a value of a jumping frequency derived from FFT, so when obtaining the value of W (f), <| E |²The value corresponding to the frequency of> must be calculated.
Thus, W is a vector having a real value for each frequency. As a special case, if W = 1 over all frequencies, equation (34) means a mere sum of powers.
[0060]
Organizing equation (34) using equations (21)-(24), (28)
P = ΣW {<| X |²> | G + HL_IL_OC |²+ <| N |²>> (35)
It becomes.
[0061]
§3-3. Silence goal
The goal of silencing is to reduce the value of equation (35). <| N |²If you focus on the point that> does not change,
Q = ΣW <| X |²> | G + HL_IL_OC |²            (36)
Can be considered to be smaller.
Note that the silence target can be subdivided according to the presence or absence of this weighting term. This point shows various cases in the calculation procedure of the silencing filter in §5.
[0062]
§4. System measurement method
§4-1. Noise wave propagation systems G and L_IL_OHow to find C
Next, the noise wave propagation systems G and L appearing in Equation (36)_IL_OHow to obtain C will be described.
Now, the characteristic of the muffler filter 7 is H₁And H₂The transfer function calculation unit 8 performs transfer function R for the two cases₁And R₂Is measured. Then, from equation (31),
R₁= G + H₁L_IL_OC (37)
R₂= G + H₂L_IL_OC (38)
Therefore, solving this simultaneous equation,
G = (H₂R₁-H₁R₂) / (H₂-H₁(39)
L_IL_OC = (R₂-R₁) / (H₂-H₁(40)
Can be obtained.
§4-2. <| X |²> How to find
<| X |²For>, if the nature of the noise source does not change over time, it is determined in advance by measuring <| X |²> May be given. Alternatively, in order to use the latest property, the transfer function calculator 8 calculates for the equation (30) <^*UU> (ie, <| U |²>) May be used. L_IThe low-pass filter 5 mainly affects the low-pass filter 5, but the low-pass filter 5 is set so as to be in the pass frequency band within the frequency range targeted for active noise control (that is, hardly attenuated). Since it is usual to design, <| U |²> As it is <| X |²It is also reasonable to use it in place of>.
[0063]
Further, in terms of correcting that the attenuation of the low-pass filter 5 gradually increases as the frequency approaches the Nyquist frequency,
1 / | L_I｜²<| U |²<| X |²It is also reasonable to use it in place of>.
In this case, 1 / | L as the muffling target frequency is exceeded and the Nyquist frequency is approached._I｜²Is approaching infinity, so at such frequencies 1 / | L_I｜²It is necessary to take measures such as stopping the multiplication.
[0064]
Then transfer function R₁, R₂<| U | when calculating in the order of²> Value to <| U₁｜²>, <| U₂｜²>, <| X |²> For later <| U₂｜²> Is also reasonable, and from the viewpoint of averaging time changes,₁｜²> And <| U₂｜²It is also reasonable to use the average of>.
§5. Calculation procedure of mute filter
§5-1. Σ | G + HC |²Mute filter to minimize
Next, the calculation procedure of the silencing filter will be described. The goal of silencing is to reduce Q in Equation (36), and a term W other than H on the right side, <| X |²>, G, L_IL_OC is known.
[0065]
First, Σ | G + HC |²The calculation method of the mute filter that minimizes the noise will be described.
Now, let the impulse responses of the noise wave propagation systems G, H, and C be g, h, and c. Considering that the multiplication in the transfer function is a convolution operation (denoted by the symbol ◎) in the time domain, G + HC is expressed as g + h ◎ c in the time domain. Then, if the FFT score when converting g + hｈc to G + HC is K, according to Percival's theorem,
Σ | G + HC |²/ K = Σ | g + h ◎ c |²                (41)
Holds.
[0066]
In the equation (41), Σ on the left side represents the sum in the frequency domain, and Σ on the right side represents the sum in the time domain. From equation (41), Σ | G + HC |²Instead of minimizing Σ | g + h ◎ c |²It can be seen that the silencing filter coefficient h that minimizes the noise may be obtained.
The muffler filter that minimizes Equation (41) can be obtained by calculating a general inverse matrix.
[0067]
Now, so that g + h ＋ c included in the right side of equation (41) can be represented by a calculation formula AZ-B using a matrix, a matrix in which c is arranged is A, a one-dimensional matrix of −g is one-dimensional of B, and h Let the matrix be Z.
The mute filter is p order, and c = {c₀, C₁, C₂, ...}, g = {g₀, G₁, G₂, ...}, h = {h₀, H₁, ..., h_p-1}
[0068]
[Expression 1]

[0069]
It is. Then, calculate the sum of the squares of each component of the matrix²The right side of equation (41) is ‖AZ-B‖²It is expressed as
Transpose the matrix A^TIf we denote A, the general inverse of A is (^TAA)^{-1 T}A,
Z = (^TAA)^{-1 T}AB (42)
‖AZ-B‖²The nature of general inverse matrix that minimizes is known (for example, Noboru Ito, Akira Iwai, Nagayoshi Iwahori, Tatsuharu Kamibayashi, Satoshi Sekino, Shuichi Takahashi “Matrix for Economics and Engineering and its Application ”, Kinokuniya, pp. 71-72 (1987)).
[0070]
If the unknown number Z is obtained, this represents the coefficient sequence h to be set in the silence filter 7.
In addition, if H expressed in the frequency domain is required for Expressions (39) and (40), it can be calculated by performing FFT on h.
Instead of calculating equation (42) directly for efficient calculation,
(^TAA) Z =^TAB (43)
Z can also be obtained by solving the following simultaneous equations. Let the autocorrelation function of c be r_iThen, on the left side (^TThe element of i row j column of AA) is r when i ≧ j._ij, If i <j_jiIt is. (^TSince AA) is a symmetric Toeplitz matrix (the same symmetric matrix with the elements arranged in the diagonal direction from the upper left to the lower right), the Levinson algorithm can be used.
[0071]
Also, on the right side^TAB is a cross-correlation function between c and -g.
The autocorrelation function of c is^*You may calculate efficiently with CC IFFT (^*C is a conjugate complex number of C).
The cross-correlation function between c and g is^*You may calculate efficiently by IFFT of CG. Needless to say, when IFFT is used, it is not necessary to obtain the impulse responses c and g once. §5-2 and §5-3, which will be described later, use this property.
[0072]
  Note that when the impulse responses g and c are first obtained, the calculation of the silencing filter coefficient h by the equation (43) itself is, for example, Yoichi Suzuki, Taichi Asano, Toshio Sone, Propagation system simulation (Part 2), Journal of the Acoustical Society of Japan, Vol. 45, No. 1, p. 44 (1989).
  In addition, a known improved technique related to the calculation of the mute filter coefficient h, for example,^TAdd diagonal element σI to AAThe(Σ is a constant, I is a unit matrix)^TWhen the determinant of AA has a small value, the value of the coefficient h can be prevented from becoming abnormally large (Toshiro Ohga, Yoshio Yamazaki, Yutaka Kaneda, "Acoustic System and Digital Processing", IEICE, p. 155 (1995)) can be used in the present invention.
[0073]
§5-2. ΣW <| X |²> | G + HC |²Mute filter to minimize
  Next, the weighting W <| X |²> Included
  ΣW <| X |²> | G + HC |²
A method for calculating a muffler filter that minimizes noise will be described.
  W <| X |²> = M²  (M is a non-negative real number). Then
  W <| X |²> | G + HC |²  = | (MG) + H (MC) ｜²
And
  ^*(MC) (MC) =^*CCM²  ,
  ^*(MC) (MG) =^*CGM²
Holds. Therefore, W <| X |²> To include the weight before IFFT described at the end of §5-1.^*With CC^*Multiply CG by this weight,^*CCW <| X |²> And^*CGW <| X |²>
[0074]
  Needless to say, W or <| X |²If only one of the weights> is included,^*With CCW^*CGW,^*CC <| X |²> And^*CG <| X |²> May be IFFT.
§5-3. ΣW <| X |²> | G + HL_IL_OC |²Mute filter to minimize
  And ΣW <| X | on the right side of Expression (36)²> | G + HL_IL_OC |²In order to calculate the silencing filter coefficient h that minimizes_IL_OC may be given. The procedure for obtaining the silence filter coefficient h in this way is hereinafter referred to as “silence filter calculation procedure”.
[0075]
§6. Start-up procedure of active noise control device
Next, using Equation (39) and Equation (40), the noise wave propagation system G and L_IL_OThe startup process of the active noise control apparatus when C is obtained will be described.
In the initial state, when there is no prior knowledge about the noise wave propagation system G and the silencer propagation system C, H₁And H₂2 ways (H₁≠ H₂) About R₁And R₂Without obtaining Eq. (39), Eq. (40) cannot be calculated. H₁And H₂One side (H for simplicity₁Can be set to 0, that is, no sound is emitted, but the remaining (H₂) Anyway, you have to put out the sound wave.
[0076]
For this reason, for example, it is determined that an extinguishing sound having an allowable size is generated, and an impulse waveform having such a size a is expressed as a coefficient h.₂It is good to assign to. For example, 1 (the output of the mute filter is output as it is) or 0.1 (a signal that is 0.1 times the input of the mute filter is output from the mute filter) is assigned as the magnitude a. H₂Is the coefficient h₂It is obtained by applying FFT to. H₁And H₂Can be other values, but H₁= 0, H₂In the case of = a, the division calculation shown in the equations (39) and (40) can be simply performed. The noise wave propagation systems G and L are expressed by Equation (39) and Equation (40)._IL_OAfter C is obtained, the silencing filter coefficient h is obtained by the above-described silencing filter calculation procedure, and if this is set in the silencing filter 7, the formula (36) which is the silencing target can be minimized.
[0077]
§7_.Maintaining silence
§7-1. System deviation detection
Next, when the noise wave propagation system G and the silencer propagation system C change during the silencing operation, a procedure for maintaining the silencing following the change will be described. For that purpose, R is always observed and L_IL_OAssume that the values of C, G, h, and H are constantly updated.
The initial value is L_IL_OC₁, G₁, H₁, H₁(The subscript starts from 1 for convenience, but this H₁Is used for startup processing₁= 0) Even when there is no system shift, the update may always be performed. Here, from the viewpoint of eliminating unnecessary updating, a case where a system shift is detected and updating is performed only when the shift is large is shown. The procedure is shown below.
(I) The mute filter is H as the initial state.₁(The coefficient is h₁), R₁Ask for.
[0078]
If there is no deviation in the system, R₁Is G₁+ H₁L_IL_OC₁Should be equal to . If the deviation is larger than a predetermined threshold value, the process proceeds to §7-2 (ii). A system shift is also a mute shift. For example, Σ | R₁-(G₁+ H₁L_IL_OC₁) ｜²Judgment is based on the value of.
§7-2. Changes in the silencer filter to deal with system shifts
(Ii) The coefficient of the silence filter is h₂Change to R₂Ask for.
[0079]
R₁-(G₁+ H₁L_IL_OC₁) Indicates a difference in mute. Which of the noise wave propagation system G and the silencer propagation system C is deviated depends on R₁It cannot be decided only from. Which is likely to shift depends on the configuration of the acoustic system. Therefore, it is determined which is likely to fluctuate, the one that is less likely to fluctuate is left as it is, and the one that is likely to fluctuate is determined as R.₁Temporarily ask for.
[0080]
Factor h that will be more useful for noise reduction₂The following three values (1) to (3) are shown as values of.
(1) If you think that the noise wave propagation system G tends to shift,
G = R₁-H₁L_IL_OC₁                          (44)
And Silencer filter coefficient h₂The noise wave propagation systems G and C₁Is calculated according to the silencing filter calculation procedure. H₂Is h by FFT₂I will ask for it.
(2) If it is considered that the silencer propagation system C tends to shift,
L_IL_OC = (R₁-G₁) / H₁                    (45)
And Silencer filter coefficient h₂Is this L_IL_OC and noise wave propagation system G₁Is calculated according to the silencing filter calculation procedure. H₂Is h by FFT₂I will ask for it.
(3) On the other hand, H₂-H₁If there is a frequency where = 0, the previously identified value must be referred to as it is at that frequency. From this point of view, H₂-H₁Since it is desirable that there is no frequency at which = 0, the impulse waveform (having power over the entire frequency range) is considered in consideration of the fact that it can be easily calculated.₁In addition to h₂A case of realizing the above will also be described. The impulse waveform is added by R₁And L_IL_OC₁Cross-correlation function (W <| X |²Is the point at which the absolute value of>

And As a result, h₂Can be obtained by applying FFT to it.₂Ask for.
[0081]
  In accordance with the explanation in §5-3, the formula (46) Appears in the molecule "R"₁And L_IL_OC₁  Cross-correlation function (W <| X |²> With weight)"Is^*(L_IL_OC₁) RW <| X |²  > Is obtained by IFFT, and “L” appears in the denominator of Equation (46)._IL_OC₁Autocorrelation function (W <| X |²> With weight)^*(L_IL_OC₁) (L_IL_OC₁) W <| X |²> Is obtained by IFFT.
[0082]
In addition, the denominator “L” in equation (46)_IL_OC₁Autocorrelation function (W <| X |²The value at time 0 with a weight of> is W <| X |²> | L_IL_OC₁｜²Is equal to the value averaged over the entire frequency range, and may be calculated accordingly.
[0083]
§7-3. System re-identification
(Iii) G₁And C₁G₂And C₂Update to
First, (ii) H₂R against₂Ask for. And R₁= G + H₀L_IL_OC and R₂= G + H₁L_IL_OFrom C, the current G and C are calculated using equations (39) and (40).
Next, ΔH = | H at each frequency₂-H₁｜²G and C depending on the size of G₁And C₁, G₂And C₂Update to As the updating method, the following two methods (1) and (2) are shown.
[0084]
(1) Qualitatively, if ΔH is large, the newly obtained G and C can be trusted. If ΔH = 0, no update is performed. In order to distinguish the difference of ΔH, a predetermined constant is α, and if ΔH ≧ α, all updates are performed. If 0 <ΔH <α, it is updated at a rate proportional to ΔH. That is, using the function min (,) that selects the smaller of the arguments, the value r = min (ΔH / α, 1) is calculated,
G₂= (1-r) G₁+ RG (47)
L_IL_OC₂= (1-r) L_IL_OC₁+ RL_IL_OC (48)
Update with.
[0085]
In the range of 0 ≦ ΔH ≦ α,

Thus, division can be avoided.
[0086]
In the range of ΔH ≧ α,
G₂=^*(H₂-H₁(H₂R₁-H₁R₂/ H₂-H₁｜²(51)
L_IL_OC₂=^*(H₂-H₁) (R₂-R₁/ H₂-H₁｜²(52)
It becomes. Therefore, in actual calculation, G and C are not calculated as intermediate results.^*(H₂-H₁(H₂R₁-H₁R₂)When^*(H₂-H₁) (R₂-R₁) Is appropriate.
[0087]
▲ 2 ▼ H₂-H₁Divide by, but there are other ways to do it. This corresponds to the magnitude of the deviation in §7-1 (i)
ΔR = | R₁-(G₁+ H₁L_IL_OC₁) ｜²Or
ΔR = | R₂-(G₁+ H₂L_IL_OC₁) ｜²When,
ΔH = | H₂-H₁｜²
G and C are updated in consideration of the size of. Qualitatively, if ΔR is large, the old C₁And G₁Is not reliable. G and C with large ΔH are reliable.
[0088]
Accordingly, α and β are predetermined constants for distinguishing the difference between ΔR and ΔH, and r = min ((αΔR + βΔH), 1) is obtained as a parameter considering both ΔR and ΔH.
G₂= (1-r) G₁+ RG (53)
L_IL_OC₂= (1-r) L_IL_OC₁+ RL_IL_OC (54)
Update with.
§7-4. Recalculate mute filter
(Iv) G₂And C₂The coefficient h of the silencing filter is calculated by the silencing filter calculation procedure from_ThreeAsk for. By FFT, h_ThreeTo H_ThreeAsk for. The coefficient h of this silence filter_ThreeUse to continue muting. That is, L_IL_OC₂, G₂, H_Three, H_ThreeEach L_IL_OC₁, G₁, H₁, H₁And return to (i).
§8. Band limitation of mute target frequency
Next, frequency band restrictions will be described. For silencer filter, L_IL_OThere is a characteristic that corrects the reverse. L_IL_OIs supposed to have a characteristic that drops in a frequency band that does not need to be muted._IL_OIt would be an unreasonable effort to correct the reverse.
[0089]
The frequency band of the speaker of the secondary sound source 2 is limited.
Now, let F be a transfer function that represents the frequency range to be silenced. The amplitude of F is set to 1 in the range where it is desired to mute, and the phase is set to 0. For example, if it is desired to mute a range of 50 to 2000 Hz, referring to the amplitude characteristics of a fourth-order BPF (bandpass filter), the amplitude of F is expressed as a function of the frequency f [Hz].

And
[0090]
In order to include this in the silence filter, H may be obtained using GF instead of G in the calculation of the silence filter calculation procedure. Since G has a value of a jumping frequency derived from FFT, when the value of F is obtained by equation (55), a value corresponding to the frequency of G must be calculated.
§9, calculation procedure of mute effect
Finally, the procedure for calculating the silencing effect will be described. The magnitude of the noise is represented by P in the equation (34), but P when no sound is extinguished._OFFAnd P when the sound is turned off_ONIn this ratio, the mute effect is shown to the user.
[0091]
Furthermore, in consideration of the general point of dB (digibel) display in the acoustic field, the silencing effect P_{ON / OFF}Is
P_{ON / OFF}= 10log_Ten(P_ON/ P_OFF(56)
It becomes. Paying attention to the fact that H = 0 should be set when not using the silencer, and expanding equation (56) using equation (35),

It can be calculated by
[0092]
As described earlier in §4-2, <| U |²> As it is <| X |²When used in place of>, (1-γ on the right side of Equation (33)²) <| D |²> As it is <| N |²It is reasonable to use it instead of>.
Further, in order to correct gradually increasing attenuation of the low-pass filter 5 as the frequency approaches the Nyquist frequency, 1 / | L_I｜²<| U |²<| X |²When used in place of>, 1 / | L_I｜²(1-γ²) <| D |²<| N |²It is reasonable to use it instead of>.
[0093]
  If the distribution of the noise power spectrum is known to be flat, <| X |²> Is a value that does not depend on the frequency, so in formula (56), <| X |²The processing of multiplying> may be omitted.
  If the frequency weighting W is flat, that is, always 1 regardless of the frequency, the formula (57It goes without saying that the process of multiplying by W can be omitted.
[0094]
<Implementation of the invention>
FIG. 4 shows the configuration of an apparatus for performing active noise control according to the present invention.
Reference numeral 21 is a reference microphone for detecting a noise wave from the noise source 20, and 22 is a speaker serving as a secondary sound source for emitting a silenced sound. The residual wave remaining after the noise wave and the sound cancellation are canceled by the 23 error microphones.
A signal from the reference microphone 21 becomes a signal x through a microphone amplifier 24, a low-pass filter 25 for anti-aliasing, and an AD converter 26, and is supplied to a DSP (digital signal processor) 27.
[0095]
The signal y output from the DSP 27 is supplied to the secondary sound source speaker 22 via the DA converter 29, the low-pass filter 30 for smoothing, and the power amplifier 31.
The signal from the error microphone 23 passes through the microphone amplifier 32, the anti-aliasing low-pass filter 33, and the AD converter 34 to become a signal e, which is supplied to the DSP 27.
[0096]
The sampling clock supply circuit 28 supplies the sampling clock to the AD converters 26 and 34 and the DA converter 29 and also supplies this sampling clock to the interrupt terminal of the DSP 27.
In the DSP 27, an interrupt routine is started in synchronization with the conversion of the AD converters 26 and 34 and the DA converter 29.
[0097]
The mute filter 7 shown in FIG. 3 is implemented in the interrupt routine of the DSP 27 in FIG. 4, and the transfer function calculation unit 8 and the adaptive control unit 15 shown in FIG. 3 are implemented in the main routine of the DSP 27 in FIG. The
In FIG. 4, for simplicity of explanation, L caused by a low-pass filter or the like is used._IAnd L_OIs considered to be included in the noise wave propagation system G and the silencer propagation system C. Hereinafter, the signal x that has passed through the AD converter 26 will be referred to as a noise wave, the signal y that is supplied to the DA converter 29 will be referred to as a muffler, and the signal e that has passed through the AD converter 34 as a residual wave. Then, the path from x to e is the noise wave propagation system G, and the path from y to e is the silencer propagation system C.
[0098]
As described above in §1 of the active noise control formulation, the formula shown in the formulation is L_I= L_OIf it is set to = 1, it will return to the Example of FIG.
In addition, when there is a propagation (feedback) of the silenced sound from the secondary sound source to the reference microphone (see FIG. 5 (a)), if the transfer function of this propagation system is set to B, as shown in FIG. 5 (b). By setting the characteristic of the silencer propagation system to C-GB, it can be reduced when there is no feedback. That is, the present invention can be applied regardless of the presence or absence of feedback.
[0099]
FIG. 6 schematically shows the function sharing between the main routine and the interrupt routine performed in the DSP 27.
In the main routine 71, transfer function calculation and silencing filter coefficient calculation are performed. In the interrupt routine 72, the noise wave and residual wave from the AD converters 26 and 34 are read, the calculation for creating the sound deadening, and the sound deadening to the DA converter 29 are written out.
[0100]
In order to exchange data between the main routine 71 and the interrupt routine 72, there are a data transfer area A3, an area A5, and an area A7. Specifically, the main routine 71 writes a new mute filter coefficient in the area A3. When new data is prepared in the area A3, the interrupt routine 72 is collectively transferred to the area A4 reserved exclusively for the interrupt routine 72.
[0101]
Since the area A4 is used for the calculation of the sound extinction in the interrupt routine 72, there is no problem in the calculation of the sound extinction even while the main routine 71 is sequentially writing data in the area A3.
Noise wave and residual wave data are written from the interrupt routine 72 to areas A5 and A7. The noise wave and residual wave data are used in the calculation of the transfer function by the cross spectrum method. When the necessary number of data is collected in the areas A5 and A7, the main routine 71 transfers the data to the areas A6 and A8. . Thereafter, the main routine 71 advances the calculation of the transfer function using the area A6 and the area A8.
[0102]
Since the calculation of the transfer function in the main routine 71 proceeds using the areas A6 and A8, even if the interrupt routine 72 writes to the areas A5 and A7 in parallel, there is no problem in calculating the transfer function.
FIG. 7 is a flowchart of interrupt processing of the DSP 27.
S10 and S11 are processes for moving the coefficient of the new silence filter to the area A4 when the coefficient is prepared in the area A3 by the main routine.
[0103]
S12 is the noise wave x_nAnd residual wave e_nAre read from the AD converter 26 and the AD converter 34, respectively. Here, the subscript n means data at the current sampling time, and ni means data at the time before the i sampling period.
S14 is the sound deadening y_nIs a process of calculating It is assumed that there are p coefficients of the mute filter.
[0104]
S15 is mute y_nIs written to the DA converter 29.
8 and 9 show the start-up process in the main routine of the DSP 27. FIG.
In S20 of FIG. 8, all work areas are initialized.
S21 is h as a coefficient of the silence filter₁Is a process for setting.
[0105]
The interrupt is permitted in S22, and thereafter, the interrupt process of FIG. 7 is started every sampling cycle.
S23 is a process for calculating a transfer function or the like by the cross spectrum method.
S24 and S25 are mute filter coefficients h₂Instead of this, a transfer function or the like is calculated. During the period until the actual silencing operation is started, 0 is set as the coefficient of the silencing filter in S26 so as not to emit the silencing sound.
[0106]
Transfer function G of noise wave propagation system in S27₁And transfer function C of the acoustic wave propagation system₁And the coefficient h of the muffler filter is calculated in S28._ThreeIs calculated.
The silence volume is calculated in S29 and S30 in FIG. 9, and displayed to the user in S31. At this time, | X₂｜², G₁, C₁, H_Three, H_Three, Γ₂ ², G₁(G₁Obtained if IFFT is applied)₁(C₁Any or all of (obtained when IFFT is applied) may be displayed.
[0107]
In S32, the user inputs whether to mute or give up. If the mute operation is performed, the process proceeds from S33 to S34, and thereafter, the process proceeds to the mute maintenance of FIG.
It should be noted that, instead of asking the user whether or not a silencing operation is possible in S32, if a silencing effect better than a predetermined threshold can be obtained, the process can automatically proceed to S34.
[0108]
FIG. 10 shows a subroutine for calculating a transfer function by the cross spectrum method called from S23 and S25 in FIG. 8 and S57 and S64 in FIG.
In S40, a work area for obtaining the sum for averaging is initialized.
In S41, areas A5 and A7 for receiving data from the interrupt routine are initialized. In S42, the process waits until the necessary number of data is collected for FFT, and in S43, the data is moved to work areas A6 and A8 dedicated to the main routine.
[0109]
A Hanning window is applied in S44, and a spectrum is calculated by FFT in S45.
S46 is a process for obtaining the sum of the power spectrum and the cross spectrum. In S47 and S48, the averaging count k is calculated.₀Iterates only and averages k times₀If so, the average is calculated by division in S49.
Thereafter, in S50, the transfer function R and the coherence function γ²Is calculated.
[0110]
  Average number of times k₀Is k in S20 of FIG.₁Set toS69 in FIG. 13 and S60c in FIG.Each in k₂And k₁Changed to k₂<K₁If the magnitude relationship is₀Is k₁In the case of, the measurement time is long but the accuracy is improved, and the number of averaging times₀Is k₂In this case, the accuracy becomes worse, but the measurement time is shorter.
[0111]
FIG. 11 shows a subroutine for calculating the silencing filter called from S28 of FIG. 8, S66 of FIG. 13, S72 of FIG. 14A, and S74 of FIG. 14B.
In S51, the autocorrelation function r (r = {r_0,r_1,r_2,... r_k-1}), A cross-correlation function d (d = {d_0,d_1,d_2,... d_k-1})). At this time, the weighting coefficient W is also included.
[0112]
After S52, the calculation of the coefficient h of the silence filter by the Levinson algorithm is shown (the number of stages of the silence filter is p).
The Levinson algorithm is described in, for example, Takaaki Tanibe, “Theory of Digital Signal Processing, 1 Basics, System, and Control”, Corona, p. 44 (1985), and so will not be described here.
[0113]
12 and 13 show processing for maintaining the mute in the main routine of the DSP 27. FIG.
S55 is the coefficient h of the silence filter₁Is set in area A3, and this coefficient h₁This is a process for starting the silencing operation by. For later processing, the coefficient h in S56₁H in the frequency domain₁Has been converted.
[0114]
Transfer function G of noise wave propagation system₁And transfer function C of the acoustic wave propagation system₁And H₁The transfer function is G₁+ H₁C₁However, this and the transfer function R obtained in S57₁Q is obtained in S58, and comparison with the threshold value δ is performed in S59. If the deviation q is equal to or less than the threshold value δ, and the average number of times k of the cross spectrum method is determined in S60c.₀Is k₁If so, since the mute state is maintained well, the state is maintained as it is, and the process proceeds to S60b aimed at saving current consumption.
[0115]
In S60b, the DSP 27 is set in the power-down mode except when interrupt processing is performed.
The process through S60b is a fixed time that can be expected to remain small in the muffled state, and is 10 seconds, for example, depending on the configuration of the acoustic system.
Averaging count k in S60c₀Is k₁Not (ie, k₂In step S60c, the averaging count k₀K₁Increase to (k₂<K₁), The measurement accuracy by the cross spectrum method is increased, and then the measurement by S57 is performed again.
[0116]
If the shift q is larger than the threshold δ in S59, the transfer function G of the noise wave propagation system₁And transfer function C of the acoustic wave propagation system₁Since the update is necessary, the process proceeds to S61.
The coefficient h determined in S61₁The coefficient h of the silence filter different from₂Is set in S62, and a transfer function corresponding to this is obtained in S64.
G at S65₁And C₁The coefficient h of the mute filter to be used next time in S66_ThreeIs calculated.
[0117]
  In S67, h so that the processing from S55 to S66 can be repeated._Three, G₂, C₂  H₁, G₁, C₁Is assigned to.
  Through S68 (and S69) and returning to S55, the process of maintaining the mute state is repeated.
  In S68 and S69, if the deviation q is larger than the threshold value δ, the averaging count k in S69.₀K₂Reduced to (k₂<K₁) By that, crossesBaeThe measurement time when performing S57 and S64 by the Kuttle method again is shortened. This is for shortening the repetition time for maintaining the mute state and improving the followability to the fluctuation of the system.
[0118]
FIG. 14 shows the coefficient h of the noise reduction filter performed in S61 of FIG.₂The three determination processes are shown. Which of the transfer function G of the noise wave propagation system and the transfer function C of the silencer propagation system is liable to shift may be determined from any of these depending on the point depending on the configuration of the acoustic system.
FIG. 14 (a) shows the transfer function G of the noise wave propagation system.₁Is assumed to have shifted to G, the coefficient h in S72₂Is shown (§7-2 (ii) (1)).
[0119]
FIG. 14B shows the transfer function C of the silencer propagation system.₁Is assumed to have shifted to C, the coefficient h in S74₂Is shown (§7-2 (ii) (2)).
FIG. 14 (c) shows the transfer function G of the noise wave propagation system.₁Is assumed, but the coefficient h₁Appropriate coefficient h under the condition that only one impulse waveform is added to₂(§7-2 (ii) (3)) is obtained.
[0120]
FIG. 15 is called G from S65 of FIG.₂And C₂The first calculation process (§7-3 (iii) (1)) is shown. S81 is G₁And C₁Is a loop for performing processing for each complex number that constitutes. In S81
H₁= {H₁ ⁽⁰⁾, H₁ ⁽¹⁾, ..., H₁ ^(k-1)}
H₂= {H₂ ⁽⁰⁾, H₂ ⁽¹⁾, ..., H₂ ^(k-1)}
R₁= {R₁ ⁽⁰⁾, R₁ ⁽¹⁾, ..., R₁ ^(k-1)}
R₂= {R₂ ⁽⁰⁾, R₂ ⁽¹⁾, ..., R₂ ^(k-1)}
G₁= {G₁ ⁽⁰⁾, G₁ ⁽¹⁾, ..., G₁ ^(k-1)}
C₁= {C₁ ⁽⁰⁾, C₁ ⁽¹⁾, ..., C₁ ^(k-1)}
G₂= {G₂ ⁽⁰⁾, G₂ ⁽¹⁾, ..., G₂ ^(k-1)}
C₂= {C₂ ⁽⁰⁾, C₂ ⁽¹⁾, ..., C₂ ^(k-1)}
It shall be expressed as In S85, Formula (51) and Formula (52) are implemented, and in S86, Formula (47) and Formula (48) are implemented.
[0121]
If (c) is adopted in FIG. 14, ΔH becomes a value unrelated to the subscript i, so S82 can be output before the loop of S81.
16 is called G from S65 in FIG.₂And C₂The second calculation process (§7-3 (iii) (2)). H in S94₂-H₁When the denominator becomes 0 in the division of, the calculation is difficult. Therefore, the processing of FIG. 14 (c) is performed in S61 of FIG. 12, and it is preferable to proceed to S61 only when there is a deviation in S59 of FIG.
[0122]
In S91,
ΔR = | R₁-(G₁+ H₁C₁) ｜²
Instead of
ΔR = | R₂-(G₁+ H₂C₁) ｜²
May be calculated.
[0123]
<Correspondence with Claimed Invention>
Finally, the correspondence between the configuration of the claimed invention and the flowchart is shown.
(1) The invention described in claim 1
Different coefficients h for the silencer filter₁And coefficient h₂The means for setting is implemented in S55, S61, and S62 of FIG.
Coefficient h₁And coefficient h₂R corresponding to each case₁And R₂The means for measuring is implemented in S57 and S64 of FIG.
[0124]
  H₁And R₁, H₂And R₂The means for calculating G and C by solving the simultaneous equations is implemented in S65 of FIG.
  New silence filter coefficient h in accordance with the least squares standard so that the silencing effect is best._ThreeTheGWhenCThe means for calculating using is implemented in S66 of FIG.
[0125]
Set different coefficients of the mute filter during the mute operation, measure the transfer function from the reference sensor signal to the error microphone signal, transfer function of the noise wave propagation path and transfer function of the mute wave propagation path Is repeated in a loop from S55 in FIG. 12 to S67 in FIG. 13. .
[0126]
(2) Invention of Claim 2
G and C and <| X |²The means for calculating> is implemented in S55 to S57 and S61 to S65 in FIG.
A means for calculating a new silence filter coefficient h using G and C in accordance with the least-squares standard so that the silencing effect is the best.²Means having a weight of> are implemented in FIG. Especially <| X |²The weighting of> is performed by multiplication in S51.
[0127]
The silencing operation using the calculated new silencing filter coefficient h is performed by setting the silencing filter coefficient in S55 after S66 and S67 in FIGS.
[0128]
(3) Invention of Claim 3
The means for calculating G and C is implemented in S55 to S57 and S61 to S65 in FIG.
The means for calculating the coefficient h of a new silence filter using G and C according to the least squares standard so that the silencing effect is the best is shown in FIG. In particular, weighting of W is performed by multiplication in S51.
The silencing operation using the calculated new silencing filter coefficient h is performed by setting the silencing filter coefficient in S55 after S66 and S67 in FIGS.
(4) Invention of Claim 4
The means for calculating G and C is implemented in S55 to S57 and S61 to S65 in FIG.
[0129]
The means for calculating the coefficient h of the new silence filter using G and C according to the least square criterion so that the silencing effect is the best while adding the weight of F to G is implemented in FIG. The weighting is performed by multiplication in S51.
The silencing operation using the calculated new silencing filter coefficient h is performed by setting the silencing filter coefficient in S55 after S66 and S67 in FIGS.
[0130]
(5) Invention of Claim 5
The value of R expected from the value of G and C stored with the coefficient h of the muffler filter (R₀Is calculated at S58 in FIG.₁+ H₁This is done by calculating C.
The value of R when the coefficient h is set in the silence filter and the silence operation is performed (R₁Is measured in S57 of FIG.
R₁And R₀The means for comparing the deviation with a predetermined threshold is implemented in S59 of FIG.
The means for calculating G and C is implemented in S55 to S57 and S61 to S65 in FIG.
[0131]
Coefficient h of new silence filter using G and C_ThreeThe means for calculating is implemented in S66 of FIG.
R₁And R₀G and C are recalculated only when the deviation is larger than a predetermined threshold, and the coefficient h of the silence filter is calculated using the recalculated G and C._ThreeAnd the coefficient h of this new silence filter_ThreeThe silencing operation using is performed on the route from S61 to S67 to S55 after branching to NO in S59.
[0132]
(6) Invention of Claim 6
Silencer filter coefficient h₁R value (R₁Is measured in S57 of FIG.
Remember C₁And H₁And R₁The means for calculating the transfer function G of the propagation path of the noise wave from the reference sensor to the error microphone is implemented in S71 of FIG. 14 (a).
[0133]
G and C₁To a new silence filter coefficient h₂The means for calculating is implemented in S72 of FIG.
New coefficient m of the calculated silence filter₂Performing the mute operation using is performed in S62 of FIG.
[0134]
(7) Invention of Claim 7
Silencer filter coefficient h₁R value (R₁Is measured in S57 of FIG.
Remembered G₁And H₁And R₁From the secondary sound source, the means for calculating the transfer function C of the propagation path of the silencer from the error microphone to the error microphone is implemented in S73 of FIG. 14 (b).
G₁And a new silence filter coefficient h from C and C₂The means for calculating is implemented in S74 of FIG. 14 (b).
[0135]
New coefficient m of the calculated silence filter₂Performing the mute operation using is performed in S62 of FIG.
[0136]
(8) Invention of Claim 8
Silencer filter coefficient h₁R value (R₁Is measured in S57 of FIG.
Remember C₁And h₁And R₁To h₂So that the silencing effect is improved.₁By adding an impulse waveform to h₂The means for calculating is implemented in S75 to S78 in FIG.
New coefficient m of the calculated silence filter₂Performing the mute operation using is performed in S62 of FIG.
[0137]
(9) Invention of Claim 9
Different coefficients h for the silencer filter₁And coefficient h₂The means for setting is implemented in S55, S61, and S62 of FIG.
Coefficient h₁And coefficient h₂R corresponding to each case₁And R₂The means for measuring is implemented in S57 and S64 of FIG.
[0138]
Silencer filter coefficient h₁, G₁, C₁Expected R value and R₁ΔR, which is the deviation of the noise, or the coefficient h of the silence filter₂, G₁, C₁Expected R value and R₂The means for calculating ΔR, which is a deviation of the above, is implemented in S91 of FIG.
H₁And R₁, H₂And R₂The means for calculating G and C by solving the simultaneous equations is implemented in S94 of FIG.
[0139]
Calculated G, C and previously stored G₁, C₁G should be used to calculate the coefficient of the muffler filter₂, C₂The means for calculating is implemented in S95 of FIG.
New silence filter coefficient h in accordance with the least squares standard so that the silencing effect is best._ThreeG₂And C₂The means for calculating using is implemented in S66 of FIG.
[0140]
When ΔR is smaller than a preset value, the calculated G and C and the previously stored G₁, C₁Is divided internally in relation to a large ΔR, so that G₂, C₂The feature of calculating is implemented in S93 and S95 in FIG.
G₂, C₂New silence filter coefficient h calculated using_ThreePerforming the mute operation using is performed in S66 of FIG.
[0141]
(10) The invention of claim 10
Different coefficients h for the silencer filter₁And coefficient h₂The means for setting is implemented in S55, S61, and S62 of FIG.
Coefficient h₁And coefficient h₂R corresponding to each case₁And R₂The means for measuring is implemented in S57 and S64 of FIG.
[0142]
  H₁And R₁, H₂And R₂G and C are calculated by solving simultaneous equations forhandThe stage is implemented by S65 in FIG. 12, that is, a combination of S83 and S85 or a combination of S83 and S86 in FIG.
  Calculated G, C and previously stored G₁, C₁G should be used to calculate the coefficient of the muffler filter₂, C₂Is calculated in S65 of FIG. 12, that is, in FIG..
  New silence filter coefficient h in accordance with the least squares standard so that the silencing effect is best._ThreeG₂And C₂The means for calculating using is implemented in S66 of FIG.
[0143]
ΔH = | H₂-H₁｜²Is smaller than the preset value, the calculated G and C and the previously stored G₁, C₁Is internally divided by the magnitude of ΔH, G₂, C₂The feature of calculating is implemented in S86 after the determination in S84 of FIG.
ΔH = | H₂-H₁｜²Is larger than a preset value, the calculated G and C are changed to G₂, C₂The feature of using for is implemented in S85 after the determination in S84 of FIG.
[0144]
G₂, C₂New silence filter coefficient h calculated using_ThreePerforming the mute operation using is performed in S66 of FIG.
The silencing operation using the calculated new silencing filter coefficient h is performed by setting the silencing filter coefficient in S55 after S66 and S67 in FIGS.
[0145]
(11) The invention of claim 11
G, C and γ²The means for calculating is implemented in S21 to S25 and S27 in FIG.
The means for calculating H based on G and C is implemented by S28 in FIG. 8 and S29 in FIG.
[0146]
G, C and γ²The means for calculating the muffled sound volume using H and H is implemented in S30 of FIG.
The means for displaying the calculated volume reduction to the user is implemented in S31 of FIG.
[0147]
(12) The invention of claim 12
Means for obtaining the transfer function R from the signal of the reference sensor to the signal to the error microphone by the cross spectrum method, in which the number of times of averaging in the cross spectrum method is variable, is implemented from S40 to S50 in FIG. ing.
The means for calculating the degree of mute shift is implemented in S58 of FIG.
The means for calculating the coefficient h of the muffler filter based on the measured R is implemented in S66 of FIG.
When the deviation of muffling is small, measuring R by performing the cross spectrum method with a larger number of averaging times means that the number of averaging times is k in S60c through S59 in FIG.₁This is implemented by performing S57 and S64.
[0148]
When there is a large amount of noise reduction, measuring the R by performing the cross spectrum method with a smaller number of averages proceeds from S68 to S69 in FIG.₂This is implemented by performing S57 and S64.
Based on the measured R, the coefficient h of the silencing filter is calculated, and the silencing operation using the calculated coefficient h is performed by performing S55 after S66 of FIG. 13 and S55. Yes.
[0149]
【The invention's effect】
  (1) According to the invention described in claim 1, since the optimum coefficient of the muffler filter is calculated from the transfer function expressed in the frequency domain without performing asymptotic iteration, the desired filter length. As a result, the follow-up performance to system fluctuations is also improved.
  Since the optimum coefficient of the muffler filter can be calculated at a desired filter length, it is a simple calculation factor as in the conventional active noise control device.TogetherThus, there is no need to increase the filter length or increase the size of the acoustic system.
[0150]
Since the calculation of the coefficient of the muffler filter can be performed in a short time, it is short to recalculate the coefficient of the muffler filter by recalculating the transfer function of the noise propagation system and the cancellation sound propagation system during the mute operation. Can be repeated in time. In the conventional active noise control apparatus, the acoustic system is limited to a fixed one such as a duct. However, the active noise control apparatus according to the present invention can follow the fluctuation of the acoustic system in a short time and can maintain the silence. For this reason, it is possible to realize a new usage form in which, for example, an error microphone is arranged on the head of a person and the sound is silenced following the movement of the person.
[0151]
  (2) According to the invention described in claim 2, it is possible to mute in consideration of the spectrum distribution of the noise wave. Therefore, the entire frequency rangeOverIn noise waves that are not always flat, more powerful frequencies can be silenced and the overall silencing effect can be enhanced.
  (3) According to the invention described in claim 3, it is possible to mute the frequency weighted. Therefore, the entire frequency rangeOverIf it is not always flat weighting, for example, noiseButIn the case of evaluation with an A-weighted sound level meter, it is possible to perform more appropriate noise reduction according to the weighting of the frequency.
[0152]
(4) According to the invention described in claim 4, it is possible to limit the frequency range to be silenced. Due to the structure of the device, it is not necessary to silence the sound at a frequency where the sound can not be emitted. In other words, since the sound is controlled to be silenced only at the frequency where the sound can be silenced, there is no excessive or wasteful occurrence.
(5) According to the invention described in claim 5, it is possible to detect system fluctuations. Since unnecessary re-measurement can be avoided when there is no system fluctuation, it can be used for power saving.
[0153]
(6) According to the invention described in claim 6, when the noise wave transfer function fluctuates, the system is measured again while increasing the silencing effect. Therefore, in an acoustic system in which the noise wave transfer function is likely to fluctuate, there is no adverse effect when the system is measured again.
(7) According to the invention described in claim 7, when the silencer transfer function fluctuates, the system is remeasured while increasing the silencing effect. Therefore, in an acoustic system in which the silence transfer function is likely to fluctuate, there is no adverse effect when the system is remeasured.
[0154]
(8) According to the invention described in claim 8, the system can be re-measured with easy post-processing. h₂-H₁Is an impulse waveform, and the frequency component covers the entire frequency band, so there is no frequency that cannot be measured. H₂-H₁Is an impulse waveform, h₂-H₁It is also possible to perform the operation divided by on the time axis. That is, since there are many options for solving the simultaneous equations, it is easy to select a convenient process when realizing the apparatus.
[0155]
(9) According to the invention described in claim 9, the system can be calculated in accordance with the degree of system fluctuation. When the system fluctuation is small, the previously measured result is taken into account, so the accuracy is high. When the system fluctuation is large, the newly measured result is emphasized and the followability is high. In other words, the balance between the followability and the high accuracy can be matched to the system situation.
(10) According to the invention described in claim 10, the system can be calculated in consideration of the calculation accuracy of the system. The previous calculation results are taken into account according to the accuracy of calculating a new system. That is, it is possible to prevent a decrease in accuracy in a certain calculation due to variations in calculation accuracy.
In addition, when calculating G and C, H₂-H₁H is almost zero, H₂-H₁Since the calculation divided by can be avoided, the calculation is not hindered.
[0156]
(11) According to the invention described in claim 11, the silencing effect is quantitatively displayed. That is, since the silencing effect is displayed using the measurement result of the system in the apparatus, the user can give up silencing in an unreasonable acoustic system, and the effort is not wasted. Also, the user can change the configuration of the acoustic system (for example, change the position of the secondary sound source) so as to obtain a more silencing effect.
[0157]
(12) According to the invention of the twelfth aspect, the measurement accuracy of the system is adjusted according to the degree of fluctuation of the system. That is, when the system fluctuation is small, the measurement takes time to increase the accuracy, and when the system fluctuation is large, the accuracy is inferior but the measurement time is lost. That is, the balance between high accuracy and followability can be matched to the system situation.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a conventional general active noise control apparatus.
FIG. 2 is a diagram showing a configuration of a conventional active noise control device for reducing noise in a passenger compartment of an automobile.
FIG. 3 is a diagram showing the configuration of an active noise control system for explaining the formulation of active noise control according to the present invention.
FIG. 4 is a diagram showing a configuration of an active noise control apparatus according to the present invention.
FIG. 5 is a diagram for explaining that when there is a feedback of silence from the secondary sound source to the reference microphone, the system can be reduced to a system without feedback.
FIG. 6 is a diagram schematically showing the function sharing of a main routine and an interrupt routine performed in the DSP 27.
FIG. 7 is a flowchart for explaining interrupt processing of the DSP 27;
FIG. 8 is a flowchart for explaining start-up processing of the present invention in the main routine of the DSP 27;
FIG. 9 is a flowchart (continuation of FIG. 8) for explaining start-up processing of the present invention in the main routine of the DSP 27;
10 is a flowchart showing a subroutine of the present invention for calculating a transfer function by the cross spectrum method, which is called from S23 and S25 in FIG. 8 and S57 and S64 in FIG.
11 is a flowchart showing a subroutine for calculating a silencing filter called from S28 of FIG. 8, S66 of FIG. 13, S72 of FIG. 14A, and S74 of FIG. 14B.
FIG. 12 is a flowchart for explaining the processing of the present invention for maintaining the mute in the main routine of the DSP 27;
FIG. 13 is a flowchart (continuation of FIG. 12) for explaining the processing of the present invention for maintaining the mute in the main routine of the DSP 27;
14 is a coefficient h of a noise reduction filter performed in S61 of FIG.₂It is a flowchart explaining the determination process of this invention.
FIG. 15 is called G from S65 of FIG.₂And C₂5 is a flowchart showing the first calculation process (§7-3 (iii) (1)).
FIG. 16 is called G from S65 of FIG.₂And C₂6 is a flowchart showing the second calculation process (§7-3 (iii) (2)).
[Explanation of symbols]
1 Reference microphone
2 Secondary sound source
3 Error microphone
4 Microphone amplifier
5 Low-pass filter
6 AD converter
7 Mute filter
8 Transfer function calculator
9 DA converter
10 Low-pass filter
11 Power amplifier
12 Microphone amplifier
13 Low-pass filter
14 AD converter
15 Adaptive control unit
20 Noise sources
21 Reference microphone
22 Speaker
23 Error microphone
24 Microphone amplifier
25 Low-pass filter
26 AD converter
27 DSP (digital signal processor)
28 Sampling clock supply circuit
29 DA converter
30 Low pass filter 30
31 Power amplifier 31
32 Microphone amplifier 32
33 Low-pass filter
34 AD converter

Claims (12)

By detecting a noise wave or a signal highly correlated therewith with a reference sensor, passing the detected signal through a silencing filter having a coefficient h, creating a silencing sound, and releasing the silencing sound from the secondary sound source, In an active noise control device that reduces noise waves after mute (hereinafter referred to as “residual waves”) at the position of the error microphone,
The transfer function of the noise wave propagation path from the reference sensor to the error microphone is represented by G, the transfer function of the muffler propagation path from the secondary sound source to the error microphone is represented by C, and the coefficient h of the mute filter is represented by the transfer function. H, where R is the measurement of the transfer function from the reference sensor signal to the error microphone signal,
Means for setting different coefficients h ₁ and h ₂ in the silencer filter;
Means for measuring R ₁ and R ₂ corresponding to the respective cases of coefficients h ₁ and h ₂ ;
Means for calculating G and C by solving simultaneous equations for H ₁ and R ₁ , H ₂ and R ₂ ;
A means for calculating a new silence filter coefficient h ₃ using G and C in accordance with the least squares standard so that the silencing effect is best;
Set different coefficients of the mute filter during the mute operation, measure the transfer function from the reference sensor signal to the error microphone signal, transfer function of the noise wave propagation path and transfer function of the mute wave propagation path An active noise control device characterized by repeatedly calculating a coefficient of a new silencing filter and performing a silencing operation using the coefficient of the silencing filter. A noise wave or a signal highly correlated with the noise is detected by a reference sensor, and the detected signal is passed through a silencer filter to create a silenced sound. In an active noise control device that reduces residual waves,
The transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, the transfer function of the propagation path of the sound wave from the secondary sound source to the error microphone is C, and the spectrum distribution of the noise wave signal is <| X | ² > And when
Means for calculating G, C, and <| X | ² > and a new muffling filter coefficient h using G and C in accordance with the least-squares standard so that the silencing effect is the best. Means for giving a weight of | X | ² >,
An active noise control device that performs a silencing operation using the calculated coefficient h of a new silencing filter. A noise wave or a signal highly correlated with the noise is detected by a reference sensor, and the detected signal is passed through a silencer filter to create a silenced sound. In an active noise control device that reduces residual waves,
G is the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone, C is the transfer function of the propagation path of the sound wave from the secondary sound source to the error microphone, and W is the frequency weighting when evaluating the silencing effect. When
Means for calculating G and C;
A means for calculating a coefficient m of a new silence filter using G and C according to the least squares standard so that the silencing effect is the best, and a means for giving a weight of W to the least squares standard. An active noise control apparatus characterized by performing a silencing operation using a coefficient h of a new silencing filter. A noise wave or a signal highly correlated with the noise is detected by a reference sensor, and the detected signal is passed through a silencer filter to create a silenced sound. In an active noise control device that reduces residual waves,
The transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, the transfer function of the propagation path of the sound wave from the secondary sound source to the error microphone is C, and the frequency weighting indicating the frequency range to be silenced is F. and when,
Means for calculating G and C;
Means for calculating a new silence filter coefficient h using G and C according to the least-squares standard so that the silencing effect is best while adding weighting of F to G;
An active noise control device that performs a silencing operation using the calculated coefficient h of a new silencing filter. By detecting a noise wave or a signal highly correlated with this with a reference sensor, passing the detected signal through a silencing filter whose coefficient is h, creating a silencing sound, and emitting the silencing sound from the secondary sound source, In an active noise control device that reduces residual waves at the position of the error microphone,
G is the transfer function of the propagation path of the noise wave from the reference sensor to the error microphone, C is the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone, and the transfer function from the signal of the reference sensor to the signal of the error microphone When R is the measured value of
The value R of R expected from the value of G and C stored with the coefficient h of the silence filter. Means for calculating
Means for measuring the value R ₁ of R when the coefficient h is set in the silence filter and the silence operation is performed;
Means for comparing the difference between R ₁ and R ₀ with a predetermined dark value;
Means for calculating G and C;
Means for calculating a new silence filter coefficient h ₃ using G and C;
Only when the difference between R ₁ and R ₀ is larger than a predetermined threshold value, G and C are recalculated. The recalculated G and C are used to obtain the coefficient h _{3 of the} silence filter, and this new active noise control apparatus which is characterized in that the mute operation using the coefficients h ₃ of silencing filter. By detecting a noise wave or a signal highly correlated with this with a reference sensor, passing the detected signal through a silencing filter whose coefficient is h, creating a silencing sound, and emitting the silencing sound from the secondary sound source, In an active noise control device that reduces residual waves at the position of the error microphone,
The transfer function of the propagation path of the muffler from the secondary sound source to the error microphone is C, the coefficient h of the mute filter is expressed as a transfer function H, and the transfer function from the reference sensor signal to the error microphone signal is measured. When the thing is R,
Means for measuring a value R ₁ of R when a silencing operation is performed with a coefficient h ₁ of the silencing filter;
Means for calculating the transfer function G of the propagation path of the noise wave from the reference sensor to the error microphone from the stored C ₁ , H ₁ and R ₁ ;
A means for calculating a new silence filter coefficient h ₂ from G and C ₁ ;
An active noise control device characterized in that a silencing operation is performed using the calculated coefficient h ₂ of the new silencing filter. By detecting a noise wave or a signal highly correlated with this with a reference sensor, passing the detected signal through a silencing filter whose coefficient is h, creating a silencing sound, and emitting the silencing sound from the secondary sound source, In an active noise control device that reduces residual waves at the position of the error microphone,
The transfer function of the noise wave propagation path from the reference sensor to the error microphone is expressed as G, the coefficient h of the mute filter expressed as a transfer function is H, and the transfer function from the reference sensor signal to the error microphone signal is measured. Is R,
Means for measuring a value R ₁ of R when a silencing operation is performed with a coefficient h ₁ of the silencing filter;
Means for calculating the transfer function C of the propagation path of the sound extinction from the secondary sound source to the error microphone from the stored G ₁ , H ₁ and R ₁ ;
A means for calculating a new silence filter coefficient h ₂ from G ₁ and C;
An active noise control device characterized in that a silencing operation is performed using the calculated coefficient h ₂ of the new silencing filter. By detecting a noise wave or a signal highly correlated with this with a reference sensor, passing the detected signal through a silencing filter whose coefficient is h, creating a silencing sound, and emitting the silencing sound from the secondary sound source, In an active noise control device that reduces residual waves at the position of the error microphone,
The transfer function of the propagation path of the muffler from the secondary sound source to the error microphone is C, the coefficient h of the mute filter expressed as a transfer function is H, and the transfer function from the reference sensor signal to the error microphone signal is measured. When the thing is R,
Means for measuring a value R ₁ of R when a silencing operation is performed with a coefficient h ₁ of the silencing filter;
Means for calculating the coefficient h ₂ by adding an impulse waveform to the coefficient h ₁ so as to improve the silencing effect when calculating the coefficient h ₂ from the stored C ₁ and coefficients h ₁ and R _1. Have
An active noise control device characterized in that a silencing operation is performed using the calculated coefficient h ₂ of the new silencing filter. By detecting a noise wave or a signal highly correlated with this with a reference sensor, passing the detected signal through a silencing filter whose coefficient is h, creating a silencing sound, and emitting the silencing sound from the secondary sound source, In an active noise control device that reduces residual waves at the position of the error microphone,
The transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is represented as G, the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone is represented as C, and the coefficient h of the silence filter is represented as a transfer function. H, where R is the measurement of the transfer function from the reference sensor signal to the error microphone signal,
Means for setting different coefficients h ₁ and h ₂ in the muffler filter;
Means for measuring R ₁ and R ₂ corresponding to the respective cases of coefficients h ₁ and h ₂ ;
Deviation between R value and R ₁ expected from coefficients h ₁ , G ₁ , and C _{1 of} silence filter, or R value and R ₂ expected from coefficients h ₂ , G ₁ , and C _{1 of} silence filter Means for calculating ΔR which is a deviation;
Means for calculating G and C by solving simultaneous equations for H ₁ and R ₁ , H ₂ and R ₂ ;
Means for calculating G ₂ and C ₂ to be used for calculating the coefficient of the silencer filter using the calculated G and C and previously stored G ₁ and C ₁ ;
The coefficients h ₃ of the new mute filter according to the minimum square norm as silencing effect is most good Ku has a means for calculating with the G ₂ and C _2,
When ΔR is smaller than a preset value, the calculated G and C and the previously stored G ₁ and C ₁ are internally divided in relation to the magnitude of ΔR, whereby G ₂ and C ₂ are calculated,
An active noise control apparatus characterized in that a silencing operation is performed using a coefficient h ₃ of a new silencing filter calculated using G ₂ and C ₂ . By detecting a noise wave or a signal highly correlated with this with a reference sensor, passing the detected signal through a silencing filter whose coefficient is h, creating a silencing sound, and emitting the silencing sound from the secondary sound source, In an active noise control device that reduces residual waves at the position of the error microphone,
The transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is represented as G, the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone is represented as C, and the coefficient h of the silence filter is represented as a transfer function. H, where R is the measurement of the transfer function from the reference sensor signal to the error microphone signal,
Means for setting different coefficients h ₁ and h ₂ in the muffler filter;
Means for measuring R ₁ and R ₂ corresponding to the respective cases of coefficients h ₁ and h ₂ ;
Means for calculating G and C by solving simultaneous equations for H ₁ and R ₁ , H ₂ and R ₂ ;
Means for calculating G ₂ and C ₂ to be used for calculating the coefficient of the silencer filter using the calculated G and C and previously stored G ₁ and C ₁ ;
A means for calculating a new silence filter coefficient h ₃ using G ₂ and C ₂ in accordance with the least squares standard so that the silencing effect is best;
When ΔH = | H ₂ −H ₁ | ² is smaller than a preset value, the calculated G and C and the previously stored G ₁ and C ₁ are larger than ΔH. To calculate G ₂ and C ₂ ,
When ΔH = | H ₂ −H ₁ | ² is larger than a preset value, the calculated G and C are used as G ₂ and C ₂ .
An active noise control apparatus characterized in that a silencing operation is performed using a coefficient h ₃ of a new silencing filter calculated using G ₂ and C ₂ . A noise wave or a signal highly correlated with the noise is detected by a reference sensor, and the detected signal is passed through a silencer filter to create a silenced sound. In an active noise control device that reduces residual waves,
The transfer function of the propagation path of the noise wave from the reference sensor to the error microphone is G, the transfer function of the propagation path of the silence wave from the secondary sound source to the error microphone is C, the transfer function of the silence filter is H, and the noise wave in the error microphone When the coherence function indicating the proportion of power derived from is γ ² ,
Means for calculating G, C and γ ² ;
Means for calculating H based on G and C;
Means for calculating the muffled sound volume using G, C, γ ² and H;
Means for displaying the calculated muffled sound level to a user. A noise wave or a signal highly correlated with the noise is detected by a reference sensor, and the detected signal is passed through a silencer filter to create a silenced sound. In an active noise control device that reduces residual waves,
Means for obtaining the transfer function R from the signal of the reference sensor to the signal to the error microphone by the cross spectrum method, wherein the number of times of averaging in the cross spectrum method is variable;
Means for calculating the degree of mute shift;
There is means for calculating the coefficient h of the muffler filter based on the measured R,
R is measured by performing the cross spectrum method with a larger number of averaging when the deviation of the mute is small, and R is measured by performing the cross spectrum method with a smaller number of averaging when the deviation of the mute is large.
An active noise control apparatus characterized in that a coefficient h of a silencing filter is calculated based on the measured R, and a silencing operation is performed using the calculated coefficient h.

Images