JP2008216359A - Noise suppressing device and receiver - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To suppress noise by reducing the influence on an input signal. <P>SOLUTION: The noise suppressing device which suppresses noise superposed on the input signal has a first filter output unit which inputs the input signal and generates a first filter output generated by attenuating the input signal to a first bandwidth including a first center frequency corresponding to frequency characteristics of predetermined noise, a second filter unit which inputs the first filter output and generates a second filter output by performing convolution integration between a predetermined weight coefficient and the first filter output, and a weight coefficient adjusting unit which adjusts the weight coefficient based upon the error between the first filter output and second filter output. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a noise suppressing device and a receiving device.

As a conventional noise suppression method, a spectral subtraction method has been proposed. FIG. 14 is a flowchart showing an outline of a conventional spectral subtraction method (see, for example, the prior art of Patent Document 1 shown below). First, when an audio signal on which noise is superimposed is input (S140), a power spectrum of the audio signal is obtained by performing Fourier transform on the audio signal (S141), and phase information of the audio signal is extracted (S142). ). Then, the power spectrum of the estimated noise is subtracted from the power spectrum of the audio signal (S143), and inverse Fourier transform is performed on the subtraction result while referring to the phase information of the audio signal (S144). As a result, an audio signal in which noise is suppressed is output (S145).
JP 2000-47698 A

  By the way, in the spectral subtraction method, when the noise estimation itself is difficult, or even if the noise can be estimated, the non-stationary noise outside the estimation is superimposed on the speech signal. In some cases, the power spectrum of the estimated noise is insufficiently subtracted or the power spectrum is excessively subtracted. In this case, a phenomenon is also known in which the waveform of a noise-suppressed audio signal is distorted, making it difficult to hear.

  Furthermore, when white noise that is correlated indefinitely between each frequency component is suppressed by the spectral subtraction method, the white noise is not suppressed as it is, but it is not heard and the colored noise is biased between the frequency components. The phenomenon that can be heard as is known. In addition, this colored noise is also called musical noise in which there is a graininess (feeling of strength for each sound) for each sound corresponding to each frequency component.

  As described above, in the spectral subtraction method, the noise is suppressed and the sound signal is also affected, and the original characteristics (such as sound graininess) of the sound signal may be changed. It was.

  A main present invention that solves the above-described problem is a noise suppression device that suppresses noise superimposed on an input signal. The noise suppression apparatus receives the input signal and includes a first center frequency corresponding to a predetermined frequency characteristic of noise. A first filter unit that generates a first filter output obtained by attenuating the input signal with a bandwidth of 1; the first filter output is input; and a predetermined weighting coefficient and the first filter output A second filter unit that performs a convolution integral to generate a second filter output; and a weighting factor adjustment unit that adjusts the weighting factor based on an error between the first filter output and the second filter output; It shall have.

  ADVANTAGE OF THE INVENTION According to this invention, the noise suppression apparatus and receiver which can suppress the noise which suppresses the influence which it has on an input signal can be provided.

<<<< System configuration of AM / FM receiver >>>>
FIG. 1 is a diagram showing a system configuration of an in-vehicle AM / FM receiver 100 according to the present invention. The AM / FM receiver 100 shown in FIG. 1 AD-converts an IF (Intermediate Frequency) signal obtained by frequency-converting the AM / FM modulated signal received by the receiving antenna 101, and processes after the AD conversion. Is digitized.

  The receiving antenna 101 is an antenna for receiving an AM / FM modulated signal transmitted from a transmitting antenna (not shown) such as a radio broadcast station. Actually, various noises are superimposed on the received AM / FM modulated signal. For example, random noise at the time of weak electric field reception and multipath noise generated by the influence of radio waves reflected on an obstacle in a multipath environment can be mentioned. Furthermore, there are ignition noise generated from a spark plug of an automobile engine, and pulse noise generated by mechanical operation of other electrical systems.

  The high frequency amplifier 102 is an amplifier that selects an AM / FM modulated signal having a preset reception frequency f1 from the reception signals received by the reception antenna 101 and amplifies the high frequency to convert it to an RF signal.

  The frequency converter 103 generates an oscillation signal having an oscillation frequency f2 different from the reception frequency f1 by a local oscillator, and then mixes it with an RF signal (high-frequency signal) output from the high-frequency amplifier 102 to generate a frequency component (f2- It is a circuit that generates signals of f1) and frequency components (f2 + f1). Further, the frequency converter 103 takes out an IF signal (intermediate frequency signal) having either a frequency component (f2-f1) or a frequency component (f2 + f1) by a BPF (Band Pass Filter) (not shown).

The intermediate frequency amplifier 104 adjusts the level of the IF signal extracted by the frequency converter 103 according to the electric field strength of the AM / FM modulated signal received by the receiving antenna 101.
The AD converter 105 converts the analog IF signal output from the intermediate frequency amplifier 104 into a digital IF signal.

The audio DSP 200 includes an AM / FM detection unit 300 and a noise suppression unit 400. In the following embodiment, it is assumed that the AM / FM detection unit 300 and the noise suppression unit 400 are implemented by a program executed by the DSP core of the audio DSP 200.
The AM / FM detection unit 300 performs AM / FM detection based on the IF signal supplied from the AD converter 105 and outputs an AM / FM demodulated signal S.
The noise suppression unit 400 performs noise suppression according to the present invention on the AM / FM demodulated signal S output from the AM / FM detection unit 300. Details of the noise suppression control will be described later.

The DA converter 106 DA-converts the digital AM / FM demodulated signal S ′ subjected to noise suppression by the noise suppressor 400 and outputs an analog AM / FM demodulated signal.
The low frequency amplifier 107 amplifies the AM / FM demodulated signal (low frequency signal) after DA conversion and outputs the amplified AM / FM demodulated signal to the speaker 108.

  The processor 800 controls the entire AM / FM receiver 100 by reading and executing a program stored in the ROM 810.

<<< Noise Suppression Unit: First Embodiment >>>
=== Entire block configuration ===
FIG. 2 is a diagram illustrating a block configuration of the noise suppression unit 400 according to the first embodiment of the present invention. As shown in FIG. 2, the noise suppression unit 400 includes a notch filter (also called a band stop filter) 410 and a line enhancer 430. The notch filter 410 is an embodiment of the “first filter section” according to the present invention, and the transversal filter 436 of the line enhancer 430 is an embodiment of the “second filter section” according to the present invention. It is a form.

  As shown in the frequency characteristic of FIG. 7, the notch filter 410 is a filter for giving a steep attenuation in the first bandwidth including the first center frequency corresponding to the frequency characteristic of predetermined pulse noise. Specifically, the notch filter 410 removes noise unnecessary for audio signal processing, that is, predetermined pulse noise, with respect to the AM / FM demodulated signal S, the spectrum distribution being biased near the first center frequency. Is. Note that the pulse noise has a frequency distribution different from white noise and multipath noise in which the spectrum distribution is biased toward high frequency components and the spectrum distribution is random.

  In the line enhancer 430, the AM / FM demodulated signal S (for example, a voice signal) generally varies gently and has a temporal correlation, whereas noise generally occurs randomly and temporally. A transversal filter is used to remove noise by utilizing the difference in characteristics that there is no correlation. Therefore, the line enhancer 430 is effective when i) the input signal has no temporal correlation (the spectral distribution is random), ii) the noise has a temporal correlation (the spectral distribution is not random). Does not work. For example, the pulse noise described above has a spectral distribution biased toward high frequency components, and is difficult to remove as noise by the line enhancer 430. Therefore, the noise suppression unit 400 of the present embodiment includes a notch filter 410 that can remove pulse noise that is difficult to remove by the line enhancer 430 in order to complement the function of the line enhancer 430.

  In addition, the noise suppression unit 400 of the present embodiment includes a notch filter 410 at the front stage and a line enhancer 430 at the rear stage. This is because the processing time of the line enhancer 430 is generally longer than the processing time of the notch filter 410. The noise suppression unit 400 is provided with a notch filter 410 at the previous stage so as to reduce the processing time of the line enhancer 430 while eliminating the influence of pulse noise in the line enhancer 430.

=== Detailed Configuration of Notch Filter ===
<Gray-Markel method>
As the notch filter 410 according to the present invention, a Gray-Markel method can be adopted. The Gray-Markel method is designed for any filter (high-pass filter, low-pass filter, band-pass filter, notch filter, etc.) based on an all-pass transfer function configured by cascading multiple basic lattices. It is a method to do. Hereinafter, the notch filter 410 employing the Gray-Markel method will be described in detail with reference to FIGS. 3 and 4.

  FIG. 3 is a diagram illustrating a basic lattice 4100 of the Gray-Markel method. 3 has two inputs (Wm + 1 (z), Sm (z)) and two outputs (Wm (z), Sm + 1 (z)). Here, “m” is a natural number indicating the number of stages of the basic lattice 4100, and “z” is a complex number of z transformation. The basic lattice 4100 includes an adder 4101a that adds the input Wm + 1 (z) and the output of the multiplier 4102a, an adder 4101b that adds the output of the multiplier 4102b and the output of the delay unit 4103, the output of the delay unit 4103, and the coefficient “− km ””, a multiplier 4102b that multiplies the output of the adder 4101a and the coefficient “km”, and a delay unit 4103 that delays the input Sm (z) by one sampling period.

式（１）、式（２）を用いて、つぎの式（３）のマトリクスが得られる。

つぎに、図４に示すように、２つの基本格子４１００ａ、４１００ｂを２段縦続接続した場合を考える。この場合の行列表現としては、式（３）を用いてつぎの式（４）として表現される。尚、ｋ１はｍ＝１の場合のｋｍであり、ｋ２はｍ＝２の場合のｋｍである。

ここで、Ｓ１（ｚ）はＷ１（ｚ）と同一である場合、式（４）からつぎの式（５）が得られる。

尚、式（５）におけるｄ１、ｄ２は、つぎの式（５’）のように定義した。

さらに、式（５）からつぎの式（６）が得られる。

尚、式（６）におけるＤ２（ｚ）は、つぎの式（６’）のように定義した。

式（６）は、２次のオールポール(All Pole)伝達関数を表している。一方、２次のオールパス(All Pass)伝達関数Ａ３（ｚ）は、式（７）で表現される。

オールポール伝達関数の式（６）とオールパス伝達関数の式（７）の分母同士を対比すると同一であることが分かる。ところで、Ｇｒａｙ−Ｍａｒｋｅｌ方式では、式（７）に示すオールパス伝達関数を基準としてフィルタ係数を調整するものである。このため、Ｇｒａｙ−Ｍａｒｋｅｌ方式のフィルタは、基本的には、図３に示した基本格子４１００を縦続接続し、最終段の出力を折り返して構成可能である。尚、この状態では、式（７）の分子に相当する構成が存在しないので、複数段縦続接続した基本格子４１００の各出力Ｓｍ（ｚ）を重み付けした上で総和をとる必要がある。この総和した結果が、Ｇｒａｙ−Ｍａｒｋｅｌ方式のフィルタの最終的な出力となる。 The input / output relationship of the basic lattice 4100 is expressed by the following equations (1) and (2).

The following equation (3) matrix is obtained using equations (1) and (2).

Next, as shown in FIG. 4, consider a case where two basic grids 4100a and 4100b are cascade-connected. In this case, the matrix expression is expressed as the following expression (4) using expression (3). Note that k1 is km when m = 1, and k2 is km when m = 2.

Here, when S1 (z) is the same as W1 (z), the following equation (5) is obtained from equation (4).

In addition, d1 and d2 in Formula (5) were defined like following Formula (5 ').

Further, the following equation (6) is obtained from the equation (5).

In addition, D2 (z) in Formula (6) was defined like the following formula (6 ′).

Equation (6) represents a second-order All Pole transfer function. On the other hand, the secondary All Pass transfer function A3 (z) is expressed by Expression (7).

It can be seen that the denominators of the all-pole transfer function equation (6) and the all-pass transfer function equation (7) are the same. By the way, in the Gray-Markel system, the filter coefficient is adjusted based on the all-pass transfer function shown in Expression (7). For this reason, the Gray-Markel filter can be basically configured by cascading the basic grids 4100 shown in FIG. 3 and folding back the output of the final stage. In this state, since there is no configuration corresponding to the numerator of Expression (7), it is necessary to calculate the sum after weighting each output Sm (z) of the basic lattice 4100 connected in cascade. The summed result is the final output of the Gray-Markel filter.

＜Ｇｒａｙ−Ｍａｒｋｅｌ方式によるノッチフィルタ＞
図５は、Ｇｒａｙ−Ｍａｒｋｅｌ方式による二次のノッチフィルタ４１０の構成を示した図である。 In order to show that the all-pass transfer function as shown in Equation (7) is a reference for all filters, a general second-order transfer function Hn (s) of a notch filter using a Laplace operator s A typical second-order transfer function Hb (s) of a bandpass filter using the Laplace operator s is taken as an example, and is represented by equations (8) and (9), respectively. In the equations (8) and (9), Q is a value called a quality factor indicating the sharpness of resonance of the filter.

<Notch filter by Gray-Markel method>
FIG. 5 is a diagram illustrating a configuration of a second-order notch filter 410 based on the Gray-Markel method.

  The notch filter 410 is basically configured by cascade-connecting gray-markel basic lattices 4100a and 4100b in two stages. Furthermore, the Gray-Markel system of the present embodiment assumes an adaptive filter configuration that automatically adjusts the filter coefficients α1 and β1. For this reason, in addition to the configuration of the basic grid 4100 shown in FIG. 3, the basic grids 4100a and 4100b include multipliers 4104a and 4104b, and adders 4105a and 4105b as mechanisms for adjusting the filter coefficients α1 and β1. Have.

式（１０）、（１１）を整理すると、つぎの式（１２）が得られる。

基本格子４１００ｂの入出力関係は、入力側Ｗｍ（ｚ）、Ｓｍ−１（ｚ）とし、出力側をＷｍ−１（ｚ）、Ｓｍ（ｚ）とすると、つぎの式（１３）、式（１４）で表現される。

式（１３）、（１４）を整理すると、つぎの式（１５）が得られる。

従って、式（１２）、式（１５）により、基本格子４１００ａ、４１００ｂを二段縦続接続した構成の伝達関数は、つぎの式（１６）によって表現される。尚、式（１６）は、式（６）に示すオールポール伝達関数として表現できる。

増幅部４１０６ａ、４１０６ｂと、加算部４１０７は、ノッチフィルタ出力Ｓ１を定めるための構成ブロックである。１段目の基本格子４１００ａの出力、すなわち加算部４１０１ｂの出力Ｓｍ＋１（ｚ）は、ゲインｇ１の増幅部４１０６ａにより増幅されて、加算部４１０７へと入力される。ＡＭ／ＦＭ復調信号Ｓは、ゲインｇ２の増幅部４１０６ｂにより増幅されて、加算部４１０７へと入力される。加算部４１０７は、増幅部４１０６ａ、４１０６ｂの各出力を加算した結果を、ノッチフィルタ出力Ｓ１として出力する。即ち、ノッチフィルタ出力Ｓ１は、つぎの式（１７）で表現される。
Ｓ１ ＝ ｇ１・Ｓｍ＋１（ｚ）＋ｇ２・Ｓ ・・・（１７）
ここで、式（１６）及び式（１７）により定まるノッチフィルタ４１０の伝達関数と、式（８）に示す一般的な伝達関数Ｈｎ（ｓ）と、を対比して、所望の第１の中心周波数及び所望の第１の帯域幅に応じた極と零点から、式（１６）に示すフィルタ係数α１、β１の各基準値を定めることができる。しかしながら、このようにフィルタ係数α１、β１を定めることは困難である。そこで、本実施形態のノッチフィルタ４１０は、所定の適応アルゴリズムに従ってフィルタ係数α１、β１を適応的に探索する最適化手法を採用し、フィルタ係数α１を調整するフィルタ係数調整部４５５ａと、フィルタ係数β１を調整するフィルタ係数調整部４５５ｂを具備する。 When the input side is Wm + 1 (z) and Sm (z) and the output side is Wm (z) and Sm + 1 (z), the input / output relationship of the basic lattice 4100a is expressed by the following equations (10) and (11). Expressed.

When the equations (10) and (11) are arranged, the following equation (12) is obtained.

When the input / output relationship of the basic lattice 4100b is Wm (z) and Sm-1 (z) on the input side and Wm-1 (z) and Sm (z) on the output side, the following equations (13) and ( 14).

By arranging the equations (13) and (14), the following equation (15) is obtained.

Therefore, the transfer function of the configuration in which the basic lattices 4100a and 4100b are cascaded in two stages is expressed by the following equation (16) by the equations (12) and (15). Equation (16) can be expressed as an all-pole transfer function shown in Equation (6).

The amplifying units 4106a and 4106b and the adding unit 4107 are constituent blocks for determining the notch filter output S1. The output of the first-stage basic lattice 4100a, that is, the output Sm + 1 (z) of the adder 4101b is amplified by the amplifying unit 4106a having the gain g1 and input to the adding unit 4107. The AM / FM demodulated signal S is amplified by the amplifying unit 4106b having the gain g2 and input to the adding unit 4107. Adder 4107 outputs the result of adding the outputs of amplifiers 4106a and 4106b as notch filter output S1. That is, the notch filter output S1 is expressed by the following equation (17).
S1 = g1 · Sm + 1 (z) + g2 · S (17)
Here, by comparing the transfer function of the notch filter 410 determined by the equations (16) and (17) with the general transfer function Hn (s) shown in the equation (8), a desired first center is obtained. The reference values of the filter coefficients α1 and β1 shown in the equation (16) can be determined from the poles and zeros corresponding to the frequency and the desired first bandwidth. However, it is difficult to determine the filter coefficients α1 and β1 in this way. Therefore, the notch filter 410 according to the present embodiment employs an optimization method that adaptively searches for the filter coefficients α1 and β1 according to a predetermined adaptive algorithm, a filter coefficient adjustment unit 455a that adjusts the filter coefficient α1, and a filter coefficient β1. Is provided with a filter coefficient adjusting unit 455b.

  The addition units 451a and 452a, the register 453a, and the filter coefficient adjustment unit 455a are configuration blocks for adaptively adjusting the filter coefficient α1. The filter coefficient α1 is used as a parameter for changing the first bandwidth as shown in FIG. In the present embodiment, the first bandwidth increases when the filter coefficient α1 increases, and the first bandwidth decreases as the filter coefficient α1 decreases.

The adder 451a subtracts the output Sm + 1 (z) of the adder 4101b from the AM / FM demodulated signal S. Adder 452a subtracts AM / FM demodulated signal S from the output of adder 451a to generate error E1. As shown in Expression (18), the error E1 is a signal obtained by inverting the output Sm + 1 (z) of the adder 4101b, that is, the output Sm + 1 (z) of the basic lattice 4100a.
E1 = S−Sm + 1 (z) −S = −Sm + 1 (z) (18)
The register 453a stores an adjustment step C1 when the filter coefficient adjustment unit 455a adjusts the filter coefficient α1.
The filter coefficient adjustment unit 455a receives the AM / FM demodulated signal S, the error E1 output from the addition unit 452a, and the adjustment step C1 stored in the register 453a. Then, in order to minimize the error E1, the filter coefficient α1, and thus the first bandwidth, is adjusted.

  FIG. 6 is a diagram showing a block configuration of the filter coefficient adjustment unit 455a. Note that the filter coefficient adjusting unit 455a shown in FIG. 6 embodies a simple gradient method (or also called steepest descent method) that minimizes the error E1.

The multiplication unit 4551 multiplies the AM / FM demodulated signal S and the error E1, and the multiplication unit 4552 multiplies the output of the multiplication unit 4551 and the adjustment unit C1. Then, the output of the multiplication unit 4552 is cumulatively added by the addition unit 4553 and the delay unit 4554. The filter coefficient adjustment unit 455a outputs the result of this cumulative addition as a filter coefficient α1. When the above calculation of the filter coefficient adjustment unit 455a is expressed by an equation, the following equation (19) is obtained.
α1 (t + 1) = α1 (t) + C1 · E1 · S
= Α1 (t) −C1 · Sm + 1 (z) · D1 (19)
From the second item of Equation (19), the filter coefficient α1 is adjusted to a value that minimizes the error E1, using “C1 · S” as the gradient information of the error E1. Note that “t” shown in Expression (19) represents the delay time of the delay unit 4554.

  The addition units 451b and 452b, the register 453b, and the filter coefficient adjustment unit 455b are configuration blocks for adjusting the filter coefficient β1. The filter coefficient β1 is used as a parameter for changing the first center frequency, as shown in FIG. In the present embodiment, the first center frequency decreases as the filter coefficient β1 increases, and the first center frequency increases as the filter coefficient β1 decreases.

Similar to the filter coefficient adjustment unit 455a shown in FIG. 6, the filter coefficient adjustment unit 455b has a configuration that embodies a simple gradient algorithm that minimizes the error E2. The error E2 used in the filter coefficient adjustment unit 455b is obtained by adding the AM / FM demodulated signal S in the adder 452b based on the result of adding the AM / FM demodulated signal S and the output Sm + 1 (z) of the adder 4101b in the adder 451b. The result of subtraction. The error E2 becomes the output Sm + 1 (z) itself of the adder 4101b as shown in the equation (20).
E2 = S + Sm + 1 (z) −S = Sm + 1 (z) (20)
When the arithmetic expression of the filter coefficient adjustment unit 455b is expressed, the following expression (21) is obtained.
β1 (t + 1) = β1 (t) + C2 · E2 · S
= Β1 (t) + C2 · Sm + 1 (z) · D2 (21)
From the second item of the equation (21), the filter coefficient β1 is adjusted to a value that minimizes the error E2, using “C2 · S” as the gradient information of the error E2. Note that “t” in Expression (21) represents the delay time of the delay unit 4554, as in Expression (19).

=== Detailed configuration of the line enhancer ===
FIG. 8 is a diagram showing a configuration of the line enhancer 430 according to the present embodiment. The line enhancer 430 shown in FIG. 8 includes a transversal filter 436 and weighting coefficient adjustment units (432a, 432b,...) Using an LMS (Least Mean Square) algorithm which is a kind of adaptive algorithm. Mainly composed.

The transversal filter 436 sequentially delays the notch filter output S1 using a delay line 431 in which N delay units (431a, 431b,...) Are connected in series. The outputs of the delay units (431a, 431b,...) Are multiplied by the weighting coefficients (γ1, γ2,...) In the multipliers (433a, 433b,. At 434, the sum is taken. The summed result is the output S ′ of the line enhancer 430, that is, the AM / FM demodulated signal S ′ in which noise is suppressed. When the arithmetic expression of the transversal filter 436 is expressed, the following expression (22) is obtained.
S ′ (t) = γ1 · S1 (t−1) + γ2 · S1 (t−2) +.
= Σγi · S1 (t−i) <i = 1 to N> (22)
From equation (22), it can be seen that the transversal filter 436 calculates a discrete convolution integral of the weighting coefficients (γ1, γ2,...) And the notch filter output S1.

The adding unit 435 generates an error E by subtracting the output S ′ of the line enhancer 430 from the notch filter output S1. That is, the error E (t) at time t is expressed by equation (23). The error E1 (t) is used when each weighting coefficient (γ1, γ2,...) Is adjusted by each weighting coefficient adjustment unit (432a, 432b,...).
E (t) = S1 (t) −S ′ (t)
= S1 (t) -Σγi · S1 (t-i) (23)
Each weighting coefficient adjustment unit (432a, 432b,...) Basically embodies an LMS algorithm that minimizes the mean square of the error E. Note that the LMS algorithm uses an RLS algorithm that recursively averages past errors E, but uses an instantaneous error E that is not averaged.

For example, the weighting coefficient adjustment unit 432a that adjusts the weighting coefficient γ1 multiplies the error E generated by the addition unit 435 and the adjustment step C stored in the register 439 by the multiplication unit 4321a, and further outputs the output of the multiplication unit 4321a. The multiplication unit 4322a multiplies the output S1 (t-1) of the delay unit 431a. Then, the adder 4323a and the delay unit 4324a cumulatively add the outputs of the multiplier 4322a. The weighting coefficient adjustment unit 432a outputs the result of this cumulative addition as a weighting coefficient γ1. When the arithmetic expression of the weighting coefficient adjustment unit 432a is expressed, the following expression (24) is obtained. Note that the delay time of the delay units (431a, 431b,...) In the delay line 431 is “t”, and the delay units (4324a, 4324b,...) Of the weighting coefficient adjustment units (432a, 432b,. Is expressed as “τ”.
γ1 (τ + 1) = γ1 (τ) + C5 · E (τ) · S1 (t−1)
= Γ1 (τ) + C5 · {S1 (τ) −Σγi · S1 (τ−i)} · S1 (t−1)
(24)
From the second item of equation (24), the weighting coefficient γ is adjusted to a value that minimizes the mean square of the error E.

  Meanwhile, it is known that the original AM / FM demodulated signal S generally has a temporal correlation, whereas noise generally has no temporal correlation. Therefore, the line enhancer 430 changes the weighting coefficients (γ1, γ2,...) According to the LMS algorithm in order to eliminate noise having no temporal correlation. As a result, when the spectrum is observed, only the linear spectrum is emphasized (enhancered), and other flat portions are interpreted as noise having no temporal correlation and attenuated.

<<< Noise Suppression Unit: Second Embodiment >>>
FIG. 9 is a diagram illustrating a block configuration of a noise suppression unit 500 according to the second embodiment of the present invention. As illustrated in FIG. 9, the noise suppression unit 500 includes a notch filter 510, a bandpass filter 520, and a line enhancer 530. That is, unlike the noise suppression unit 400 according to the first embodiment of the present invention, the noise suppression unit 500 of the present embodiment is provided with a bandpass filter 520 between the notch filter 510 and the line enhancer 530. It is in the point. The bandpass filter 520 is an embodiment of the “third filter unit” according to the present invention.

  In the following description, the notch filter 510 and the line enhancer 530 are the same as the notch filter 410 and the line enhancer 430 in the first embodiment of the present invention, and therefore only the bandpass filter 520 will be described.

  Since the AM / FM demodulated signal S is generally a low-frequency component of about several tens of kilohertz, it can be handled separately from the high-frequency component of pulse noise. Therefore, the notch filter 510 generates a notch filter output S1 from which pulse noise has been removed from the AM / FM demodulated signal S, and inputs the notch filter output S1 to the bandpass filter 520. Then, the bandpass filter 520 generates a bandpass filter output S2 in which the notch filter output S1 is passed with a second bandwidth including the second center frequency corresponding to the frequency characteristic of the AM / FM demodulated signal S, Input to line enhancer 530.

  As a result, the line enhancer 530 handles the AM / FM demodulated signal S in which the pulse noise is removed by the notch filter 510 and is limited to a desired frequency band from which the noise is to be removed by the band pass filter 520. it can. Thereby, the effect of noise suppression in the line enhancer 530 is further enhanced.

  FIG. 10 is a diagram illustrating a configuration of a second-order bandpass filter 520 using the Gray-Markel method.

  The band-pass filter 520 shown in FIG. 10 is basically the same as the configuration of the notch filter 410 of the first embodiment shown in FIG. 5, and the Gray-Markel basic lattices 4100c and d are connected in two stages. Configured. Filter coefficients α2 and β2 are used. The transfer function of the configuration in which the basic lattices 4100a and 4100b shown in FIG. 10 are cascaded in two stages is expressed by the following equation (25).

ところで、バンドパスフィルタ出力Ｓ２は、乗算部５２０６ａにおいてノッチフィルタ出力Ｓ１をゲインｇ３と乗算した結果と、乗算部５２０６ｂにおいて加算部４１０１ｆの出力Ｓｍ＋１（ｚ）をゲインｇ４と乗算した結果と、を加算部５２０７において減算した結果となる。式で表現すると、つぎの式（２６）となる。
Ｓ２ ＝ ｇ３・Ｓ１ − ｇ４・Ｓｍ＋１（ｚ） ・・・ （２６）
式（２５）と式（２６）とを対比して、加算部４１０１ｆの出力Ｓｍ＋１（ｚ）は互いに逆極性であり、バンドパスフィルタ５２０の特性は、ノッチフィルタ５１０の特性を裏返したものであることが分かる。ここで、式（２５）及び式（２６）により定まるバンドパスフィルタ５２０の伝達関数と、式（９）に示す一般的な伝達関数Ｈｂ（ｓ）と、を対比して、所望の第２の中心周波数及び所望の第２の帯域幅に応じた極と零点から、式（２５）に示すフィルタ係数α２、β２の各基準値を定めることができる。しかしながら、フィルタ係数α１、β１の場合と同様、フィルタ係数α２、β２それぞれの基準値を定めることは困難である。そこで、本実施形態のバンドパスフィルタ５２０は、所定の適応アルゴリズムに従ってフィルタ係数α２、β２を適応的に探索する最適化手法を採用し、フィルタ係数α２を調整するフィルタ係数調整部５２５ａ、フィルタ係数β２を調整するフィルタ係数調整部５２５ｂを具備する。

By the way, the bandpass filter output S2 is obtained by adding the result obtained by multiplying the notch filter output S1 by the gain g3 in the multiplier 5206a and the result obtained by multiplying the output Sm + 1 (z) of the adder 4101f by the gain g4 in the multiplier 5206b. The result of subtraction in the unit 5207 is obtained. Expressed as an expression, the following expression (26) is obtained.
S2 = g3 · S1−g4 · Sm + 1 (z) (26)
Comparing Expression (25) and Expression (26), the output Sm + 1 (z) of the adder 4101f is opposite in polarity, and the characteristic of the bandpass filter 520 is the reverse of the characteristic of the notch filter 510. I understand that. Here, the transfer function of the bandpass filter 520 determined by the equations (25) and (26) and the general transfer function Hb (s) shown in the equation (9) are compared, and the desired second The reference values of the filter coefficients α2 and β2 shown in Expression (25) can be determined from the pole and zero corresponding to the center frequency and the desired second bandwidth. However, as in the case of the filter coefficients α1 and β1, it is difficult to determine the reference values for the filter coefficients α2 and β2. Therefore, the band-pass filter 520 of the present embodiment employs an optimization method that adaptively searches for the filter coefficients α2 and β2 in accordance with a predetermined adaptive algorithm, a filter coefficient adjustment unit 525a that adjusts the filter coefficient α2, and a filter coefficient β2 Is provided with a filter coefficient adjustment unit 525b.

  The filter coefficient adjustment unit 525a is an embodiment of the “second bandwidth variable unit” according to the present invention. The filter coefficient α2 adjusted by the filter coefficient adjustment unit 525a is used as a parameter for changing the second bandwidth as shown in FIG. In the present embodiment, the second bandwidth is increased when the filter coefficient α2 is increased, and the second bandwidth is decreased when the filter coefficient α2 is decreased.

  The filter coefficient adjustment unit 525b is an embodiment of the “second center frequency variable unit” according to the present invention. The filter coefficient β2 adjusted by the filter coefficient adjustment unit 525b is used as a parameter for changing the second center frequency, as shown in FIG. In the present embodiment, the second center frequency decreases as the filter coefficient β2 increases, and the second center frequency increases as the filter coefficient β2 decreases.

<< Effects of noise suppression unit >>
FIG. 12A is a diagram showing a waveform of an input signal on which noise is superimposed, and FIG. 12B is a diagram showing a waveform of the input signal after noise is suppressed according to the present invention. .
FIG. 13A is a diagram showing the waveform of the input signal on which noise is superimposed, and FIG. 13B shows the waveform of the input signal after the noise is suppressed by the conventional spectral subtraction method. FIG. 13C is a diagram showing the waveform of the input signal after noise is suppressed according to the present invention.
From the comparison between FIG. 12A and FIG. 12B, it can be seen that the noise superimposed on the input signal is certainly suppressed by the present invention. 13 (b) and FIG. 13 (c), the input signal is attenuated as noise is suppressed in the conventional spectral subtraction method, whereas in the present invention, the input signal is affected. It can be seen that noise is suppressed without giving.

<<< Other Embodiments >>>>
As mentioned above, although embodiment of this invention was described, embodiment mentioned above is for making an understanding of this invention easy, and is not for limiting and interpreting this invention. The present invention can be changed / improved without departing from the spirit thereof, and the present invention includes equivalents thereof.

  The system having the noise suppression units 400 and 500 according to the present invention is not limited to the AM / FM receiver 100 described above, and may be an AM receiver or FM receiver, or an optical disc recording / reproducing apparatus. Also good.

  The noise suppression units 400 and 500 according to the present invention have been described as being based on programs executed by the DSP core of the audio DSP 200, but may be implemented as analog / digital circuits that implement the functions of those programs.

  The adaptive algorithm for adjusting the weighting coefficient γi (i = 1 to N) at each stage in the line enhancers 430 and 530 according to the present invention is not limited to the LMS algorithm described above, but other algorithms (RLS (Recursive Least)). Squares) etc. may be adopted.

  Similarly, as the adaptive algorithm for adjusting the filter coefficients α1, α2, β1, and β2 in the filter coefficient adjusting units 455a, 455b, 525a, and 525b of the notch filters 410 and 510, other adaptive algorithms besides the above-described simple gradient algorithm are used. (Conjugate gradient method etc.) may be adopted.

It is the figure which showed the system configuration | structure of the AM / FM receiver based on this invention. It is the figure which showed the block configuration of the noise suppression part which concerns on 1st Embodiment of this invention. It is the figure which showed the basic grid of the Gray-Markel system which concerns on this invention. It is the figure which showed the structure at the time of two-stage cascade connection of the basic grid of the Gray-Markel system which concerns on this invention. It is the figure which showed the structure of the notch filter which concerns on this invention. It is the figure which showed the block configuration of the filter coefficient adjustment part which concerns on this invention. It is the figure which showed the frequency characteristic of the notch filter which concerns on this invention. It is the figure which showed the structure of the line enhancer which concerns on this invention. It is the figure which showed the block configuration of the noise suppression part which concerns on 2nd Embodiment of this invention. It is the figure which showed the structure of the band pass filter which concerns on this invention. It is the figure which showed the frequency characteristic of the band pass filter which concerns on this invention. (A) is the figure which showed the waveform of the input signal on which the noise was superimposed, (b) is the figure which showed the waveform of the input signal after noise was suppressed by this invention. (A) is the figure which showed the waveform of the input signal with which the noise was superimposed, (b) is the figure which showed the waveform of the input signal after noise was suppressed by the conventional spectrum subtraction method, (c) ) Is a diagram showing a waveform of an input signal after noise is suppressed according to the present invention. It is the flowchart which showed the outline | summary of the conventional spectrum subtraction method.

Explanation of symbols

100 AM / FM receiver 101 Reception antenna 102 High frequency amplifier 103 Frequency converter 104 Intermediate frequency amplifier 105 AD converter 106 DA converter 107 Low frequency amplifier 108 Speaker 200 Audio DSP
300 AM / FM detector 400, 500 Noise suppressor 410, 510 Notch filter 520 Bandpass filter 430, 530 Line enhancer 4100, 4100a, 4100b Basic lattice 4101a, 4101b, 4101c, 4101d, 4105a, 4105b, 451a, 451b, 4107, 452a, 452b, 4553, 4323a, 4323b, 434, 435, 521a, 521b, 5207, 522a, 522b, 5207 adder 4102a, 4102b, 4102c, 4102d, 4104a, 4104b, 4106a, 4106b, 4551, 4552, 4321a , 4321b, 4322a, 4322b, 433a, 433b, 5206a, 5206b multipliers 4103, 4103a, 4103b, 4554 431a, 431b, 4324a, 4324b delay unit 453a, 453b, 439,523a, 523b registers 455a, 455b, 525a, 525b filter-coefficient adjusting unit 431 delay line 800 the processor 810 ROM

Claims (9)

In a noise suppression device that suppresses noise superimposed on an input signal,
A first filter unit that receives the input signal and generates a first filter output obtained by attenuating the input signal with a first bandwidth including a first center frequency corresponding to a predetermined frequency characteristic of noise; ,
A second filter unit that receives the first filter output, performs a convolution integration of a predetermined weighting coefficient and the first filter output, and generates a second filter output;
A weighting coefficient adjusting unit that adjusts the weighting coefficient based on an error between the first filter output and the second filter output;
A noise suppression device comprising: In a noise suppression device that suppresses noise superimposed on an input signal,
A first filter unit that receives the input signal and generates a first filter output obtained by attenuating the input signal with a first bandwidth including a first center frequency corresponding to a predetermined frequency characteristic of noise; ,
The first filter output is input, and a third filter output is generated by passing the first filter output with a second bandwidth including a second center frequency corresponding to the frequency characteristic of the input signal. A third filter unit;
A second filter unit that receives the third filter output, performs a convolution integration of a predetermined weighting coefficient and the first filter output, and generates a second filter output;
A weighting coefficient adjusting unit that adjusts the weighting coefficient based on an error between the third filter output and the second filter output;
A noise suppression device comprising: The noise suppression device according to claim 1 or 2,
The first filter unit adjusts a first filter coefficient to vary the first center frequency based on an error between the input signal and the first filter output. A noise suppression device comprising a variable unit. The noise suppression device according to claim 3.
The noise suppression apparatus according to claim 1, wherein the first center frequency variable unit adjusts the first filter coefficient by an algorithm for minimizing the error. The noise suppression device according to claim 1 or 2,
The first filter unit adjusts a first filter coefficient to vary the first bandwidth based on an error between the input signal and the first filter output. A noise suppression device comprising a variable unit. The noise suppression device according to claim 2,
The third filter unit adjusts a second filter coefficient to vary the second center frequency based on an error between the first filter output and the third filter output. And a center frequency variable unit. The noise suppression device according to claim 2,
The third filter unit adjusts a second filter coefficient to vary the second bandwidth based on an error between the first filter output and the third filter output. A noise suppression device comprising: a bandwidth variable unit. The noise suppression device according to claim 1,
The weighting coefficient adjustment unit includes:
A noise suppression apparatus, wherein the weighting coefficient is adjusted by an algorithm that minimizes a mean square of the error. In a receiving apparatus that demodulates a modulated reception signal received by an antenna,
A first filter unit that receives the received signal and generates a first filter output in which the received signal is attenuated by a first bandwidth including a first center frequency corresponding to a frequency characteristic of predetermined noise; ,
A second filter unit that receives the first filter output, performs a convolution integration of a predetermined weighting coefficient and the first filter output, and generates a second filter output;
A weighting coefficient adjusting unit that adjusts the weighting coefficient based on an error between the first filter output and the second filter output;
A receiving apparatus comprising:

Images