JP3591870B2 - Active noise control system - Google Patents

Description

【００３０】
【数２】

但し、ここでｗは適応フィルタ係数、ｉはフィルタ係数の番号、μはステップサイズ、ｅはエラー記号、ｒは伝達関数補正用フィルタ出力信号、Ｃはマイクロホンとスピーカとの間の伝達関数、Ｘは参照入力信号、ｊはＣのインパルス応答の番号、ｋはＣのタップ数である。
【００３１】
（数２）がＦｉｌｔｅｒｅｄ−Ｘの特徴である空間（スピーカ〜マイクロホン間）の伝達関数Ｃと入力データＸの畳込み演算である。
【００３２】
ここで参照入力に次式のようなインパルスを入力したとする。ただし、このインパルスは消音対象ある周期性騒音に同期して繰り返される。
【００３３】
Ｘ（０）＝１
Ｘ（ｉ）＝０ ｉ＜０ ｉ＞０
上記式の条件を満たすということは、次式が成り立つ。
【００３４】
ｎ＝ｊ→Ｘ（ｎ−ｊ）＝１
ｎ≠ｊ→Ｘ（ｎ−ｊ）＝０
すると（数２）は次式になる。
【００３５】
ｒ（ｎ）＝ｃｎ
すなわち、ｒ（ｎ）は伝達関数Ｃのインパスルレスポンスｃｎ を順次出力することであり、畳込み演算を必要としない。フィルタＷに関しても、これと同様に畳込み演算をせずにフィルタ係数ｗｉを順次出力すればよく、演算量を大幅に削減できかつ、図３のような通常のＦｉｌｔｅｒｅｄ−Ｘの参照入力にパルスが入力した場合と演算結果は等価である。これがＳｙｎｃｈｒｏｎｉｚｅｄ Ｆｉｌｔｅｒｅｄ−Ｘ（ＳＦＸ）アルゴリズムである。
【００３６】
次に、一括更新型ＳＦＸ−ＬＭＳについて説明する。
【００３７】
ＳＦＸでは（数１），（数２）は、（数３）になる。
【００３８】
【数３】

従来のアルゴリズムでは図３（Ｈ部）の様に（数３）をｋ＋１回、１サンプルで更新していた。従って、フィルタＷの係数ｗｉ の更新が完了するためには、Ｃのタップ数であるｋ＋１回分更新しなければならない。この説明を図４を使い説明する。（数３）を図式化したものが図４である。縦に並ぶｅとｃが（数３）の右辺第２項の乗算を示す。これは、Ｃを８タップ（ｋ＝７）とし、参照入力のパルスが２２サンプル目に入力した場合の例であり、縦軸がサンプル数（ｎ）、横軸がフィルタＷのタップ番号（ｉ）である。Ｐ１、Ｐ２、Ｐ３はポインタの位置を示している。これらポインタについては後で説明する。
【００３９】
例えば、１１サンプル目のｗｉを更新する場合を考える。従来型のＳＦＸ−ＬＭＳアルゴリズムでは（数３）に従い、ｗ１９からｗ１２までの８個の係数を更新する（図４の点線で囲んだ部分の演算）。この処理でｗ１２の更新は終了する。
【００４０】
次にｗ１２に着目すると、このタップは、４サンプル目から更新が始まり、１１サンプル目で更新が終了する。更新が完了したｗ１２´は（数４）の様に示される。
【００４１】
【数４】

尚、（ｗｉ´）はｗｉの更新が完了していることを示す。
【００４２】
（数４）を一般式に直すと（数５）で表される。
【００４３】
【数５】

（数５）式の右辺第２項はエラー信号と伝達関数Ｃの係数ｃｊを逆並べした係数との畳込み演算である。（図４の実線で囲んだ部分の演算を示す。）（数５）は、従来の（数３）をｋ＋１回演算するのと比べ、乗算、減算ともｋ＋１回削減することができる。
【００４４】
また、ＤＳＰの特徴として、畳込み演算を得意とするアーキテクチャを採用しいるため、実際は、削減された演算数以上に、演算結果を短縮することができる。
【００４５】
これが、一括更新型ＳＦＸ−ＬＭＳアルゴリズムである。図２のＵＰＦがこの一括更新型ＳＦＸ−ＬＭＳによる更新係数演算部であり、処理としては、伝達関数Ｃを推定したＣｈａｔのインパルスレスポンスを逆並べした数列を係数とするＦＩＲフィルタである。
【００４６】
なお、図４において、Ｐ１は係数出力用ポインタである。ポインタはフィルタ係数（ｗ）上をサンプルごとにインクリメントされる。ＳＦＸなので畳込み演算を行わず、Ｐ１の示すデータｗｉを出力すればよい。また、このポインタは、参照入力にパルスが入力するとｗ０ に戻る。すなわち適応フィルタのタップ数が可変である。Ｐ２は一括更新する係数ｗを示す係数更新用ポインタである。Ｐ３はＳＴＦ出力をＡＤＳＧに格納するポインタである。
【００４７】
次にＡＤＳＧ係数安定化ＦＩＲフィルタ（ＳＴＦ）について説明する。
【００４８】
本システムの適応フィルタが不安定になる条件としては、次のようなことが考えられる。
【００４９】
１）騒音の周波数帯域をスピーカＳＰ１〜４の特性がカバーできない場合、フィルタ係数が発散する。例えば、騒音の周波数が５０Ｈｚで、スピーカＳＰ１〜４が５０Ｈｚのような低域を発生できない場合、適応フィルタは５０Ｈｚの信号を生成するが、エラー信号は減少しないために、適応フィルタが発散してしまう。
【００５０】
２）Ｆｉｌｔｅｒｅｄ−Ｘ ＬＭＳの特性として、高音域が発散しやすい。
【００５１】
３）マイクロホンＭＩＣ１〜４入力に何かの理由でオフセットがかかった場合、適応フィルタにもＤＣ成分が重畳してしまう。
【００５２】
４）エンジン回転数が変化した場合、適応フィルタの出力に不連続点が生じてしまう。
【００５３】
これらの問題点を解決する方法として、フィルタ係数安定化フィルタ（ＳＴＦ）を採用した。
【００５４】
このフィルタ係数安定化フィルタの原理について説明する。
【００５５】
一括更新型ＳＦＸ−ＬＭＳアルゴリズムにより更新される値を直線位相のＦＩＲフィルタでフィルタリングする。このＦＩＲフィルタの特性は基本的にはバンドパスフィルタとし、スピーカＳＰ１〜４の再生できない低音域と、空間で消音できない高音域の更新をカットする。このフィルタにより、低音域や高音域の発散を防止し、また、エンジン回転数が変化した場合の不連続音の発生を減少させることができる。
【００５６】
ＳＴＦに直線位相のＦＩＲフィルタを使うことで、位相状態は保たれたまま周波数のフィルタリングが可能となる。このＳＴＦによる遅延はタップ数の半分のサンプル数となるため、その時間経過後に、適応フィルタの更新を行わなければならない。
【００５７】
この説明を図４を使い説明する。ＳＴＦを７タップのＦＩＲフィルタとした場合、フィルタリングされた結果は、３サンプル後に出力される。従って、（数５）の結果であるｗｉ´（ｎ＋１）をフィルタリングした値は、３サンプル後に更新すれば良い。Ｐ３が更新されるアドレスである。
【００５８】
次に回転数、回転状態感応型ステップサイズ、ＳＴＦ変更方式について説明する。
【００５９】
ＳＦＸアルゴリズムの欠点として、その構造上、エラー信号の影響が直接出力信号に反映される。言い換えると、エラー信号に参照入力と相関性のない信号が入力した場合、出力信号にその相関性のない信号が重畳されてしまっていた。
【００６０】
本システムのようなＡＮＣの場合、エラーマイクロホンに向かって声を発する等のことを行うと、スピーカよりエコーが発生する場合がある。このような対策として、ステップサイズを小さくし、更新量を小さくすることでエコーを抑える方法があるが、この処理は、システムの性能を劣化させてしまう。
【００６１】
従来は、これらのバランスを考えてステップサイズを決定していた。その結果、エンジンの加減速に追従する性能を出すことができなかった。
【００６２】
また、もう一つの問題として、アルゴリズムの構成上、加減速時にキャンセル信号に不連続点が発生する場合がある。その対策として、ＳＴＦを付加したが、単一の特性では効果的に不連続音を低減させることができなかった。
【００６３】
そこで、本実施例においてはエンジン回転数、回転状態を監視し、これによりステップサイズ、ＳＴＦを変更することで、前記の問題点が緩和させた。
【００６４】
この方式の原理について説明する。
【００６５】
エンジン回転数、回転状態をＤＳＰを使い監視し、それにより、ステップサイズ、ＳＴＦを制御させた。回転数は低回転、中回転、高回転の３段階に、回転状態は、定常回転、加速、減速、急減速の４段階に分け、これらの組み合わせの計１２通りの場合分けを行い、夫々に最適なステップサイズ、ＳＴＦの係数を図示せぬ記憶手段であるメモリに記憶しておき、当該メモリから夫々を選択した。
【００６６】
図５にエンジン回転数、回転状態に対するステップサイズ、ＳＴＦの例を、図６にＳＴＦの周波数特性を示す。例えば、加減速時は、不連続点が発生しやすいため、ローパス・フィルタのカットオフ周波数を低く設定する。この時のカットオフ周波数は、２００〜４００Ｈｚが適当である。また騒音の変化に対する追従性を上げるため、ステップサイズを大きく設定する。
【００６７】
対して、定常回転時は、なるべく高い周波数まで消音するため、ローパス・フィルタのカットオフ周波数を高くし、エコーを抑えるために、ステップサイズを小さくする。ギアチェンジ等の急減速時は、適応処理が追従できないとして、ＡＤＳＧの係数をすべてクリアする。なお、図５、図６は説明のための例であり、この値にとらわれるものではない。
【００６８】
次にエンジンパルス割り込みによるサンプリングタイミング調整機能について説明する。
【００６９】
本システムのアルゴリズム（一括更新型ＳＦＸ−ＬＭＳ）は適応フィルタのタップ長をエンジン回転数により変化させる可変タップアルゴリズムである。適応フィルタのタップ長はエンジン二回転の周期に一致させるのであるが、これは、サンプリング周期を単位として離散的になる。従って、従来の方法では騒音の周期とキャンセル音の周期を正確に一致させることができず、ビート音を発生するという問題があった。
【００７０】
そこで、この対策として、エンジンパルスをサンプリング周期の基準とし、強制的に騒音の周期とキャンセル音の周期とを一致させる。この動作を図６に示す。
【００７１】
具体的には、エンジンパルスをＤＳＰの外部割り込みに入力し、エンジンパルスにサンプリング周期（ＡＤ，ＤＡコンバータのタイミング）を同期させる。エンジンパルスが入力されると、図７のＮ番目のサンプルの処理を中断し、新たなサンプリングを開始する。これにより、エンジンパルス入力時のみ疑似的にサンプリング周波数が高くなることになり、ビート音の発生を抑えることができる。
【００７２】
また、先に前記フィルタ係数安定化フィルタ（ＳＴＦ）の原理説明にて説明したように、いかなる場合であれ、図４に示すように係数出力用ポインタＰ１と係数更新用ポインタＰ２との間隔は、ＦＩＲフィルタ（ＵＰＦ）のタップ数分であり、係数更新用ポインタＰ２とＳＴＦの出力結果をＡＤＳＧに格納するポンイタＰ３との間隔は、前記ＳＴＦのタップ数の半分であり、このような各ポインタ間の間隔におけるＳＴＦ及ＵＰＦのタップ数における因果律は不変である。
【００７３】
また、前記係数出力用ポインタＰ１から係数を出力した結果、係数更新用ポインタＰ２の係数が算出され、当該係数更新用ポインタＰ２の係数をＳＴＦに入力した結果、ポインタＰ３にて前記算出された値がＡＤＳＧに格納される。
【００７４】
ところが、エンジンの回転数が上がると、当該エンジン回転数に同期したエンジンパルスの間隔が短くなり、係数出力用ポインタＰ１がポインタＰ３を追い抜いてしまう場合が発生する。この場合には、先に説明した各ポインタ間の間隔におけるＳＴＦ及びＵＰＦのタップ数における因果律が満たせなくなり、正確な適応が行えなくなって、ノイズが発生するといった問題があった。
【００７５】
そこで、本発明のアクティブ・ノイズ・コントロール・システムによれば、このような場合に限り、エンジンパルスを一つ無視して、つまり当該一つのエンジンパルスを読み飛ばすことにより、係数出力用ポインタＰ１がポインタＰ３を追い抜いてしまうといった事態の発生を無くすようにしたので、ノイズの発生を防止することができる。
【００７６】
また、エンジンパルス間隔で、騒音が繰り返されるので、キャンセル音も繰り返される。そこで、はじめのエンジンパルス間隔分のキャンセル信号を読み飛ばされたエンジンパルス以降に複写することにより、不連続音に対しより一層の効果を上げることができる。
【００７７】
【発明の効果】
上記のように構成された本発明のアクティブ・ノイズ・コントロールシステムによれば、急速にエンジン回転数が上がって所定間隔よりも短い間隔でエンジンパルスが入力されると、キャンセル信号を生成する処理がおぼつかなくなって、キャンセル信号に発生した不連続点によってノイズが発生するといった事態を打開することができる。
【図面の簡単な説明】
【図１】本実施例の構成を立体的に表わした斜視図である。
【図２】ＡＮＣモードの概略を示すブロック図である。
【図３】ＳＦＸ−ＬＭＳの基本的なアルゴリズムを示すブロック図である。
【図４】一括更新型ＳＦＸ−ＬＭＳを図式化した説明図である。
【図５】エンジン回転数、回転状態のチューニング結果を示す説明図である。
【図６】ＳＴＦの係数と特性を示す説明図である。
【図７】エンジンパルス割り込み状態を示す概念図である。
【符号の説明】
ＳＰ１〜４ キャンセル音出力用スピーカ
ＭＩＣ１〜４ エラーマイクロホン
Ｃ１１〜４４ 伝達関数
Ｃｈａｔ１１〜４４ 伝達関数
ＡＤＳＧ１〜４ 適応フィルタ
ＳＴＦ ＡＤＳＧ係数安定化フィルタ
ＵＰＦ 一括更新型ＳＦＸ−ＬＭＳ係数更新用ＦＩＲフィルタ
ＭＹＵ２ Ｒ ステップサイズ２
ＥＮＧＩＮＥ ＰＵＬＳＥ エンジンパルス
１０ エンジン回転数・回転状態判断部[0001]
[Industrial applications]
The present invention relates to an active noise control system (hereinafter simply referred to as ANC) for reducing engine noise of an automobile, a ship, or the like by spatially canceling the noise by using an adaptive filter.
[0002]
[Prior art]
Conventionally, as an adaptive filter for generating a cancel signal for reducing engine noise in such an ANC, a cancel signal is generated based on a reference input signal which is an engine pulse synchronized with the engine noise and an error signal extracted from a microphone. You do it.
[0003]
[Problems to be solved by the invention]
However, according to the above-described conventional ANC, the adaptive filter generates a cancel signal based on the engine pulse and the error signal. However, when the engine speed is rapidly rotated, the tap interval of the adaptive filter is reduced. When the length is shortened, the engine pulse becomes shorter than a predetermined interval, and there is a problem that a discontinuous point occurs in the cancel signal and noise occurs.
[0004]
The present invention has been made in view of the above problems, and an object of the present invention is to provide an active noise control system that prevents noise due to discontinuous points of a cancel signal generated at the time of rapid rotation of an engine. Is to do.
[0005]
[Means for Solving the Problems]
In order to achieve the above object, the present invention has an adaptive filter that generates a cancel signal with an engine pulse synchronized with engine noise, and based on the generated cancel signal, in a closed space in a vehicle compartment or the like based on the generated cancel signal. A system for canceling and reducing the engine noise, wherein, when the engine pulse is input at an interval shorter than a predetermined interval, only a specific engine pulse among a plurality of engine pulses input at the short interval. The cancel signal is generated, and the other engine pulses are skipped without generating the cancel signal.
[0006]
[Action]
With this configuration, when the engine speed is rapidly increased and the engine pulse is input at an interval shorter than the predetermined interval, the process of generating the cancel signal is not obscured, and noise is generated due to a discontinuous point generated in the cancel signal. The situation that arises can be overcome.
[0007]
【Example】
Hereinafter, an embodiment of an active noise control system according to the present invention will be described.
[0008]
An active noise control system (hereinafter, referred to as ANC) for a vehicle will be described as an example. This is a system in which an applied filter is applied to reduce the noise near the occupant's head by canceling the engine noise in space.
The signal processing characteristics of the ANC mode of the present system are listed below.
[0009]
1) The batch update SFX-LMS algorithm was devised to reduce the amount of calculation.
[0010]
2) A coefficient stabilizing FIR filter (STF) for preventing discontinuous sounds, preventing the divergence of the adaptive filter, and preventing the growth of unnecessary low-frequency sounds is employed.
[0011]
3) A rotation speed, rotation state sensitive type, STF, and step size change method are adopted.
[0012]
4) Adoption of sampling timing adjustment function by engine pulse interruption.
[0013]
FIG. 1 is a perspective view three-dimensionally illustrating the configuration of the present embodiment. In FIG. 1, an engine 100 which is a noise source of engine noise, an engine control unit 200 which detects a rotation state and the number of rotations of the engine 100, a speaker 300 installed at an arbitrary position in the vehicle, and It has a microphone 400 provided at an upper part for collecting noise in the vehicle, and an ANC controller 500 for controlling the entire ANC.
[0014]
FIG. 2 shows a block diagram of the ANC mode. SP1 to SP4 are cancellation sound output speakers, and MIC1 to MIC4 are error microphones.
[0015]
C11 to C44 are transfer functions between the speakers SP1 to SP4 and the microphones MIC1 to MIC4, and Chat11 to C44 are transfer functions between the speakers SP1 to SP4 and the microphones MIC1 to MIC4 estimated by system identification.
[0016]
“ADSG1 to 4” are adaptive digital signal generators (cancellation signal generators), which have been generally used conventionally.
[0017]
"STF" is an ADSG coefficient stabilizing FIR filter, "UPF" is a batch update type SFX-LMS coefficient updating FIR filter, "SS1" is a step size 1, a fixed step size applied to each pass, and "SS2" is a step. Size 2 is a step size that changes depending on the number of rotations and the rotation state.
[0018]
W1 (n) to W4 (n) are ADSG coefficients before updating. Reference numeral 11 denotes a phase inverting unit that inverts the phases of the output values output from the ADSGs 1 to 4. Reference numeral 10 denotes an engine speed / rotation state determination unit that is a monitoring unit and a control unit. ENGINEPULSE is a pulse synchronized with ADSG1 to ADSG4 and the engine speed input to the engine speed / rotation state determination unit 10.
[0019]
As shown in FIG. 1, this system is a CASE (1, 4, 4) using one engine 100 as a noise source and using four error microphones 400 and four speakers 300.
[0020]
Next, the operation of the block diagram shown in FIG. 2 will be briefly described.
[0021]
The sum of the engine noise and the cancel signal generated from the speakers SP1 to SP4 is captured by the microphones MIC1 to MIC4 and used as an error signal. Then, a convolution operation of the error signal and a numerical value obtained by rearranging (Data Reverse) the estimated transfer function (Chat) is performed. This is a feature of the batch update type SFX-LMS. This FIR filter is called UPF. (Details of the SFX-LMS will be described later). Further, the convolved value is multiplied by a step size of 1.
[0022]
The result of the multiplication by the step size 1 is added for each of the microphones MIC1 to MIC4 input, and the result is multiplied by a step size 2 (SS2) according to the engine speed and the rotation state. This is an engine speed and rotation state sensitive type step size changing method. The calculation result is the updated value of the ADSG coefficient.
[0023]
The updated values are respectively added to the current ADSG coefficients [W1 (n)] to [W4 (n)].
[0024]
In order to stabilize the addition result, filtering is performed by the STF, and the result is set as new ADSG coefficients [W1 (n + 1)] to [W4 (n + 1)]. The characteristics of the STF change depending on the engine speed and the rotation state. This is the rotation speed sensitive STF changing method.
[0025]
Further, the phase of the output value of the signal output from the ADSG is inverted by the phase inverting unit 11. Then, the signals are output from the speakers SP1 to SP4. The above is the basic operation of the ANC mode.
[0026]
Next, the Filtered-X LMS will be described.
[0027]
In normal Filtered-X LMS, when implementing ANC in a three-dimensional space, a signal correlated with a signal (noise) to be silenced is taken in as a reference signal. FIG. 3 shows a signal processing algorithm in this case. This is a general Filtered-XLMS algorithm.
[0028]
The equation for updating the coefficient of the Filtered-X LMS is shown in (Equation 1) and (Equation 2).
[0029]
(Equation 1)

[0030]
(Equation 2)

Here, w is an adaptive filter coefficient, i is a filter coefficient number, μ is a step size, e is an error symbol, r is a transfer function correction filter output signal, C is a transfer function between a microphone and a speaker, X Is the reference input signal, j is the number of the C impulse response, and k is the number of C taps.
[0031]
(Equation 2) is a convolution operation of the transfer function C in space (between the speaker and the microphone) and the input data X, which is a feature of Filtered-X.
[0032]
Here, it is assumed that an impulse represented by the following equation is input to the reference input. However, this impulse is repeated in synchronization with the periodic noise to be silenced.
[0033]
X (0) = 1
X (i) = 0 i <0 i> 0
Satisfying the condition of the above expression satisfies the following expression.
[0034]
n = j → X (n−j) = 1
n ≠ j → X (n−j) = 0
Then, (Equation 2) becomes the following equation.
[0035]
r (n) = cn
That is, r (n) means to sequentially output the impulse response cn of the transfer function C, and does not require a convolution operation. Similarly, for the filter W, the filter coefficient wi may be sequentially output without performing the convolution operation, so that the amount of operation can be greatly reduced, and a pulse is input to the reference input of the ordinary Filtered-X as shown in FIG. Is equivalent to the operation result. This is the Synchronized Filtered-X (SFX) algorithm.
[0036]
Next, the batch update type SFX-LMS will be described.
[0037]
In SFX, (Equation 1) and (Equation 2) become (Equation 3).
[0038]
(Equation 3)

In the conventional algorithm, (Equation 3) is updated k + 1 times with one sample as shown in FIG. 3 (part H). Therefore, in order to complete the update of the coefficient wi of the filter W, it must be updated by k + 1 times, which is the number of taps of C. This will be described with reference to FIG. FIG. 4 is a diagram of Equation (3). The e and c arranged vertically indicate the multiplication of the second term on the right side of (Equation 3). This is an example in which C is set to 8 taps (k = 7) and the reference input pulse is input at the 22nd sample. The vertical axis represents the number of samples (n), and the horizontal axis represents the tap number (i) of the filter W. ). P1, P2, and P3 indicate the positions of the pointers. These pointers will be described later.
[0039]
For example, consider a case where the wi of the eleventh sample is updated. In the conventional SFX-LMS algorithm, eight coefficients from w19 to w12 are updated in accordance with (Equation 3) (operation in a portion surrounded by a dotted line in FIG. 4). This process ends the update of w12.
[0040]
Next, focusing on w12, this tap starts updating from the fourth sample, and ends updating at the eleventh sample. The updated w12 'is shown as (Equation 4).
[0041]
(Equation 4)

Note that (wi ') indicates that the update of wi has been completed.
[0042]
(Equation 4) can be expressed by (Equation 5) when converted into a general equation.
[0043]
(Equation 5)

The second term on the right side of the expression (5) is a convolution operation of the error signal and the coefficient obtained by rearranging the coefficient cj of the transfer function C. (The operation of the portion surrounded by the solid line in FIG. 4 is shown.) (Equation 5) can reduce k + 1 times for both multiplication and subtraction compared to the conventional (Equation 3) which is operated k + 1 times.
[0044]
Further, as a feature of the DSP, an architecture that is good at convolution operation is adopted, so that the operation result can be actually shortened more than the reduced number of operations.
[0045]
This is the batch update type SFX-LMS algorithm. The UPF in FIG. 2 is an update coefficient calculation unit based on the batch update type SFX-LMS, and the processing is an FIR filter that uses a sequence of reversely arranged Chat impulse responses for which the transfer function C is estimated as coefficients.
[0046]
In FIG. 4, P1 is a coefficient output pointer. The pointer is incremented for each sample on the filter coefficient (w). Since it is SFX, the convolution operation is not performed, and the data wi indicated by P1 may be output. This pointer returns to w0 when a pulse is input to the reference input. That is, the number of taps of the adaptive filter is variable. P2 is a coefficient update pointer indicating the coefficient w to be updated collectively. P3 is a pointer for storing the STF output in ADSG.
[0047]
Next, an ADSG coefficient stabilized FIR filter (STF) will be described.
[0048]
The following are conceivable conditions for making the adaptive filter of the present system unstable.
[0049]
1) When the characteristics of the speakers SP1 to SP4 cannot cover the frequency band of the noise, the filter coefficient diverges. For example, when the frequency of the noise is 50 Hz and the speakers SP1 to SP4 cannot generate a low band such as 50 Hz, the adaptive filter generates a signal of 50 Hz. However, since the error signal does not decrease, the adaptive filter diverges. I will.
[0050]
2) As a characteristic of the Filtered-XLMS, a high-frequency range is easily diverged.
[0051]
3) When an offset is applied to the inputs of the microphones MIC1 to MIC4 for some reason, the DC component is also superimposed on the adaptive filter.
[0052]
4) When the engine speed changes, a discontinuous point occurs in the output of the adaptive filter.
[0053]
As a method for solving these problems, a filter coefficient stabilizing filter (STF) is employed.
[0054]
The principle of this filter coefficient stabilizing filter will be described.
[0055]
The value updated by the batch update type SFX-LMS algorithm is filtered by a linear phase FIR filter. The characteristics of the FIR filter are basically band-pass filters, and cut off the updating of the low sound range that cannot be reproduced by the speakers SP1 to SP4 and the high sound range that cannot be silenced in space. With this filter, it is possible to prevent divergence in the low frequency range and the high frequency range, and to reduce the occurrence of discontinuous sounds when the engine speed changes.
[0056]
By using a linear phase FIR filter for the STF, frequency filtering can be performed while the phase state is maintained. Since the delay due to the STF is half the number of samples of the number of taps, the adaptive filter must be updated after the elapse of the time.
[0057]
This will be described with reference to FIG. If the STF is a 7-tap FIR filter, the filtered result is output after three samples. Therefore, the value obtained by filtering wi ′ (n + 1), which is the result of Equation 5, may be updated after three samples. P3 is the address to be updated.
[0058]
Next, a description will be given of a rotation speed, a rotation state-sensitive step size, and an STF changing method.
[0059]
As a disadvantage of the SFX algorithm, due to its structure, the effect of the error signal is directly reflected on the output signal. In other words, when a signal having no correlation with the reference input is input to the error signal, the signal having no correlation is superimposed on the output signal.
[0060]
In the case of an ANC such as the present system, if a voice is uttered toward an error microphone, an echo may be generated from a speaker. As a countermeasure for this, there is a method of suppressing echo by reducing the step size and the update amount, but this processing degrades the performance of the system.
[0061]
Conventionally, the step size is determined in consideration of these balances. As a result, it was not possible to achieve the performance of following the acceleration / deceleration of the engine.
[0062]
As another problem, a discontinuity point may occur in the cancel signal during acceleration / deceleration due to the configuration of the algorithm. As a countermeasure, STF was added, but discontinuous sounds could not be effectively reduced with a single characteristic.
[0063]
Therefore, in the present embodiment, the above-mentioned problems were alleviated by monitoring the engine speed and the rotation state, and thereby changing the step size and the STF.
[0064]
The principle of this method will be described.
[0065]
The engine speed and the rotation status were monitored using a DSP, thereby controlling the step size and STF. The number of rotations is divided into three stages: low rotation, medium rotation, and high rotation, and the rotation state is divided into four stages: steady rotation, acceleration, deceleration, and rapid deceleration. The optimal step size and the STF coefficient were stored in a memory, which is a storage unit (not shown), and each was selected from the memory.
[0066]
FIG. 5 shows an example of the engine speed, the step size for the rotation state, and the STF, and FIG. 6 shows the frequency characteristics of the STF. For example, at the time of acceleration / deceleration, the cutoff frequency of the low-pass filter is set low because discontinuous points are likely to occur. The cutoff frequency at this time is suitably 200 to 400 Hz. In addition, the step size is set to be large in order to improve the followability to the change in noise.
[0067]
On the other hand, at the time of steady rotation, the cutoff frequency of the low-pass filter is increased to mute the sound as high as possible, and the step size is reduced to suppress echo. At the time of sudden deceleration such as a gear change, the adaptive processing cannot be followed, and all ADSG coefficients are cleared. FIGS. 5 and 6 are examples for explanation, and are not limited to these values.
[0068]
Next, a sampling timing adjustment function by an engine pulse interrupt will be described.
[0069]
The algorithm of the present system (batch update type SFX-LMS) is a variable tap algorithm that changes the tap length of the adaptive filter according to the engine speed. The tap length of the adaptive filter is made to coincide with the cycle of two engine revolutions, which is discrete with the sampling cycle as a unit. Therefore, the conventional method cannot accurately match the cycle of the noise with the cycle of the cancel sound, and has a problem that a beat sound is generated.
[0070]
Therefore, as a countermeasure, the engine pulse is used as a reference for the sampling cycle, and the cycle of the noise and the cycle of the cancel sound are forcibly made to match. This operation is shown in FIG.
[0071]
More specifically, an engine pulse is input to an external interrupt of the DSP, and a sampling cycle (timing of the AD and DA converters) is synchronized with the engine pulse. When the engine pulse is input, the processing of the N-th sample in FIG. 7 is interrupted, and a new sampling is started. As a result, the sampling frequency is artificially increased only at the time of inputting the engine pulse, and the generation of the beat sound can be suppressed.
[0072]
In any case, as described in the principle of the filter coefficient stabilizing filter (STF), the interval between the coefficient output pointer P1 and the coefficient update pointer P2 is, as shown in FIG. The number of taps of the FIR filter (UPF) is equal to the number of taps, and the interval between the coefficient update pointer P2 and the ponita P3 for storing the output result of the STF in the ADSG is half the number of taps of the STF. The causality in the number of taps of the STF and the UPF in the interval is constant.
[0073]
Further, as a result of outputting the coefficient from the coefficient output pointer P1, the coefficient of the coefficient updating pointer P2 is calculated. As a result of inputting the coefficient of the coefficient updating pointer P2 to the STF, the calculated value is calculated by the pointer P3. Is stored in ADSG.
[0074]
However, when the engine speed increases, the interval between engine pulses synchronized with the engine speed decreases, and the coefficient output pointer P1 may overtake the pointer P3. In this case, the causality in the tap numbers of the STF and the UPF in the interval between the pointers described above cannot be satisfied, so that accurate adaptation cannot be performed and noise occurs.
[0075]
Therefore, according to the active noise control system of the present invention, only in such a case, by ignoring one engine pulse, that is, by skipping the one engine pulse, the coefficient output pointer P1 Since the occurrence of the situation of overtaking the pointer P3 is eliminated, it is possible to prevent the occurrence of noise.
[0076]
Further, since the noise is repeated at the engine pulse intervals, the cancel sound is also repeated. Therefore, by copying the cancel signal for the first engine pulse interval after the skipped engine pulse, it is possible to further improve the effect of the discontinuous sound.
[0077]
【The invention's effect】
According to the active noise control system of the present invention configured as described above, when the engine speed is rapidly increased and an engine pulse is input at an interval shorter than a predetermined interval, a process of generating a cancel signal is performed. It is possible to overcome a situation in which noise is generated due to a discontinuous point generated in the cancel signal when the signal becomes unclear.
[Brief description of the drawings]
FIG. 1 is a perspective view three-dimensionally illustrating the configuration of the present embodiment.
FIG. 2 is a block diagram schematically showing an ANC mode.
FIG. 3 is a block diagram showing a basic algorithm of SFX-LMS.
FIG. 4 is an explanatory diagram schematically illustrating a batch update type SFX-LMS.
FIG. 5 is an explanatory diagram showing tuning results of an engine speed and a rotation state.
FIG. 6 is an explanatory diagram showing STF coefficients and characteristics.
FIG. 7 is a conceptual diagram showing an engine pulse interrupt state.
[Explanation of symbols]
SP1 to 4 Cancel sound output speakers MIC1 to 4 Error microphones C11 to 44 Transfer functions Chat11 to 44 Transfer functions ADSG1 to 4 Adaptive filters STF ADSG coefficient stabilization filter UPF Batch update type SFX-LMS coefficient update FIR filter MYU2 R Step size 2
ENGINE PULSE Engine pulse 10 Engine speed / rotation status judgment unit

Claims (2)

A system that has an adaptive filter that generates a cancel signal with an engine pulse synchronized with the engine noise, and that cancels and reduces the engine noise in a closed space in a vehicle compartment or the like based on the generated cancel signal. So,
When the engine pulse is input at an interval shorter than a predetermined interval, among the plurality of engine pulses input at the short interval, the cancel signal is generated only for a specific engine pulse, and for other engine pulses, the cancel signal is generated. An active noise control system characterized by skipping without generating a cancel signal. 2. The active noise control system according to claim 1, wherein a cancel signal generated immediately before the skipped engine pulse is copied to the skipped engine pulse.

Images