JP2015126360A - Signal processor, and signal processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately eliminate a noise with a periodicity in frequency characteristic from a signal.SOLUTION: Noise has a frequency characteristic whose autocorrelation varies periodically in frequency. A signal processor includes: frequency shift means for generating a signal whose frequency is shifted by an optional frequency except zero, which makes the autocorrelation maximum; equalizing means for compensating a frequency characteristic of a frequency shifted signal outputted by the frequency shift means; combining means for combining a signal outputted by the equalizing means and the input signal; and control means for controlling a transfer function of the equalizing means so as to suppress noise in the combined signal outputted by the combining means.

Description

  The present invention relates to a signal processing device and a signal processing method for removing noise from a signal including noise having periodicity in frequency characteristics.

  As a method for demodulating a desired broadcast wave by removing the noise from the desired broadcast wave on which the noise is superimposed, techniques described in the following patent documents are known. The following Patent Document 1 discloses a first antenna that receives a broadcast wave stronger than the other antenna and obtains a first signal, and a second antenna that receives noise stronger than the other antenna and obtains a second signal. Is used to adjust the amplitude and phase of the second signal so that the signal level after synthesis is reduced, and synthesizes the first signal. In this technique, the amplitude and phase of the second signal are adjusted when the reception level of the desired broadcast wave is smaller than a certain threshold value. That is, when the noise power is larger than the power of the broadcast wave, the amplitude and phase of the second signal are adjusted so that the level of the combined signal becomes small. Since both antennas receive noise from the same noise source, the amplitude and phase of the noise received by both antennas differ according to the difference in distance from each antenna to the noise source. In order to compensate for this, the amplitude of the second signal is made to coincide with the amplitude of the first signal, the phase of the second signal is changed by π with respect to the phase of the first signal, and The second signal is synthesized in reverse phase. By adjusting the amplification factor and phase of the second signal in this way, a signal with noise canceled from the received broadcast wave can be obtained even when a desired broadcast wave can be received.

  Further, in the technique of Patent Document 2 below, in receiving an AM radio broadcast wave by a radio receiver mounted on a vehicle, pulsed noise emitted from an electronic device of the vehicle is mixed in the AM radio broadcast wave. This is a technology for removing sexual noise. In this technique, first, pulse noise in a band other than the AM radio broadcast wave band is detected, and the noise source is specified by obtaining the magnitude of the period of the pulse noise and the fluctuation range of the level. Depending on the noise source, in the broadcast wave on which pulse noise is superimposed, the harmonic band of the pulse noise near the broadcast wave frequency and the band of the noise period are attenuated for the time corresponding to the time width of the pulse noise. By doing so, the person who listens to AM radio broadcasting is prevented from giving discomfort due to pulse noise.

JP2012-257155A Patent No. 5012246

  The technique of Patent Document 1 is a technique that uses two antennas so that only noise can be received and is adjusted in advance so that the level of the combined signal of the received signals of the two antennas becomes small. For this reason, the technique of Patent Document 1 requires two antennas to cancel noise, and the setting for removing noise does not affect the adjustment of noise removal by the reception level of a desired broadcast wave. Thus, it is necessary to perform in an environment where the reception level is small. Further, since the amplitude and phase of the second signal are adjusted uniformly regardless of the frequency characteristics of the noise, the noise is not completely removed. Further, in the case of a DC-DC converter using the PWM method, the frequency characteristic of noise changes due to a change in pulse width even if the basic period does not change. For this reason, the method of Patent Document 1 cannot completely remove noise.

  The technique of Patent Document 2 detects pulse noise in a band other than the broadcast wave band, and attenuates an appropriate frequency band according to the type of noise for the time corresponding to the pulse width at the detection timing. Technology. Therefore, the broadcast wave is essentially attenuated only during the period of pulse noise. This causes discomfort to those who listen to AM radio broadcasts.

  Accordingly, an object of the present invention is to accurately remove noise having periodicity in a frequency space without affecting a desired signal.

  According to a first aspect of the present invention for solving the above-described problems, in the signal processing apparatus for removing noise from an input signal in which noise is superimposed on a desired signal to be detected, the noise has an autocorrelation periodically with respect to the frequency. A frequency transition means for generating a signal whose frequency is shifted by an arbitrary non-zero frequency among the frequencies where the autocorrelation has a maximum frequency characteristic, and a frequency transition output from the frequency transition means. Equalizing means for correcting the frequency characteristics of the signal, synthesizing means for synthesizing the signal output from the equalizing means, and the input signal, and noise in the synthesized signal output by the synthesizing means at least in a band including the desired signal And a control means for controlling a transfer function of the equalization means so as to suppress the signal processing apparatus.

  In the present invention, the noise to be removed is noise having periodicity in the frequency space. Here, the term “periodicity” is a concept that includes the case where the frequency characteristic of noise is a periodic function and the case where the characteristic of each period is correlated. In other words, the frequency characteristic of noise may be a characteristic such that the autocorrelation of the frequency characteristic varies periodically with respect to the frequency. It is sufficient that the maximum value of autocorrelation exists at a plurality of frequencies. For example, when the noise is a pulse having a period T on the time axis, a line spectrum with a frequency interval of 1 / T is obtained on the frequency axis, and the autocorrelation of this frequency characteristic takes a maximum value at an interval of 1 / T. In this case, the frequency characteristic of noise has periodicity. Even when the noise is a square wave having a width τ of the period T, the frequency characteristic of the noise is not a complete periodic function, but the autocorrelation takes a maximum value at a frequency interval of 1 / T. The characteristic has periodicity. Further, the noise does not have to be a periodic function on the time axis, and the frequency characteristics may be continuous.

  Although the present invention can be constituted by an analog circuit, the input signal may be sampled and processed by a computer as a digital value. The frequency transition means is a means for performing processing for removing a product of a sine wave having a frequency to be shifted and an input signal and a lower band of the product. The frequency transition means can be realized by a mixer and a band-pass filter, or a computer program that performs the same processing.

  In the present invention, when the fundamental period of the noise source on the time axis is known, the frequency shifted by the frequency shifting means may be the known frequency. For example, if the carrier frequency of a DC-DC converter in a vehicle is 100 kHz, the frequency characteristic of the noise becomes a line spectrum at 100 kHz intervals, and the autocorrelation takes a maximum value at 100 kHz intervals. Therefore, it can be said that the frequency characteristic of the noise generated from the DC-DC converter has periodicity. In this case, the frequency transition amount of the input signal is an integral multiple of 100 kHz.

  Further, the present invention has transition frequency determining means for determining the autocorrelation of the frequency characteristic of the input signal, determining the maximum value of the autocorrelation, and determining a non-zero frequency that takes the maximum value, The input signal may be shifted in frequency by the frequency determined by the transition frequency determining unit. That is, the periodicity in the frequency characteristics of noise may be measured from the input signal. For example, in the case of the above-described DC-DC converter that operates at 100 kHz, the autocorrelation of the frequency characteristic of the input signal takes a maximum value at intervals of 100 kHz. When the operating frequency changes or the pulse width changes, the frequency characteristics of the input signal change. However, by using the maximum value of the autocorrelation as the amount of frequency transition, noise is suppressed as described later. it can.

  In the present invention, the transition frequency determining means is a Fourier transform means for obtaining the frequency characteristic of the input signal, a non-zero frequency that obtains the autocorrelation of the frequency characteristic obtained by the Fourier transform means, and takes the maximum value of the autocorrelation. It is good also as a means which has an autocorrelation calculating means to determine. This is one configuration for specifically obtaining the autocorrelation of the frequency characteristic of the input signal. The transition frequency determination means may obtain the autocorrelation of the frequency characteristic of the input signal from the square frequency characteristic of the input signal. This utilizes the fact that the autocorrelation of a function is equal to the square Fourier transform of the function.

  In the present invention, the equalizing means may be means for outputting a convolution of the impulse response of the transfer function (inverse Fourier transform of the transfer function) and the frequency transition signal. Further, the equalizing means may be a means for performing a transversal filter process in which a transfer function is realized by a sequential delay between taps and a weighting factor that multiplies a signal delayed and branched by each tap. This is a specific example of performing convolution of the frequency transition signal and the impulse response of the transfer function. The transversal filter may be an analog circuit or may be executed by a calculation process by a computer.

  In the present invention, the equalizing means may be means for obtaining a frequency characteristic of the frequency transition signal and outputting an impulse response of a product of the frequency characteristic and the transfer function. This process is performed in the frequency space, and the frequency characteristic of the frequency transition signal is corrected by the inverse Fourier transform of the product of the Fourier transform of the frequency transition signal and the transfer function.

  In the present invention, the control means may be means for determining a transfer function by a power inversion algorithm that minimizes the power of the combined signal. That is, the desired signal is included in the synthesized signal, but when the power of noise is large, minimizing the power of the synthesized signal is equivalent to removing the noise to the maximum. Further, the control means may be means for determining a transfer function so as to minimize an error between the synthesized signal and the known reference signal. That is, determining the transfer function so that the synthesized signal becomes a known reference signal is equivalent to increasing the amplification factor of only the desired signal, and noise that has no correlation with the reference signal is suppressed from the synthesized signal. Is done. If the frequency is shifted by the frequency at which the autocorrelation takes a maximum value in the frequency characteristics of noise, the noise is overlapped according to the magnitude of the autocorrelation due to the similarity of characteristics. As a result, if the frequency characteristics of the frequency transition signal are modified so that the power of the combined signal is minimized or the reference signal is increased and synthesized with the frequency characteristics of the input signal, noise with autocorrelation is obtained. Only components can be removed. In other words, the input signal and the frequency transition signal can be synthesized so as to remove noise having periodicity in frequency characteristics.

  Although the present invention does not limit the desired signal or noise source, the desired signal is an AM radio broadcast wave, and the noise is, for example, noise generated from a DC-DC converter or an inverter in a hybrid vehicle. Therefore, the AM radio broadcast can be heard with high quality in the vehicle.

The principle of the present invention can also be understood as the following method invention.
In the method invention, the noise has a frequency characteristic in which the autocorrelation periodically varies with respect to the frequency, and the input signal is shifted in frequency by an arbitrary non-zero frequency among the frequencies at which the autocorrelation is maximized. Generate a frequency transition signal, modify the frequency characteristics of the frequency transition signal, synthesize the modified frequency transition signal and the input signal, and suppress noise in the synthesized signal at least in a band including the desired signal A signal processing method characterized by correcting the frequency characteristics of the frequency transition signal.
Also in the present invention, the above explanation is valid as it is.

  According to the present invention, noise having frequency characteristics with periodicity can be accurately removed from an input signal, so that detection accuracy and demodulation accuracy of a desired signal can be improved.

The block diagram of the signal processing apparatus which concerns on the specific Example 1 of this invention. The frequency characteristic of the input signal input into the signal processing apparatus of Example 1. Frequency characteristics of the frequency transition signal obtained by shifting the input signal by the frequency f _d . The frequency characteristics of the synthesized signal. The timing chart which showed the correlation matrix, the timing which calculates a weight vector, the timing of weighting addition, and the timing of a signal sequence.

  Hereinafter, the present invention will be described based on specific examples. The present invention is not limited to the following examples.

  A configuration of a signal processing apparatus 1 according to a specific embodiment of the present invention is shown in FIG. A present Example is a signal processing apparatus which suppresses the noise mixed in the AM radio receiver in HV (hybrid vehicle). It is assumed that the HV is equipped with a DC-DC converter that is controlled at a carrier frequency of 100 kHz. The AM radio broadcast wave is assigned a frequency band of 531 kHz to 1602 kHz. The switching noise generated from the DC-DC converter basically becomes a line spectrum string that is an integral multiple of the fundamental frequency of 100 kHz in the frequency space. This noise enters the AM radio broadcast band and gives noise to the AM radio broadcast wave. The present embodiment is a signal processing apparatus that cancels this noise entering the AM radio broadcast band.

In the signal processing apparatus 1 of this embodiment, the AM radio broadcast signal received by the antenna 11 is sampled by the A / D converter 13 at a constant period Δt and converted into a digital value, and then processed by the CPU. Device. The configuration of FIG. 1 is expressed in blocks for each functional unit of digital processing. An input signal S ₁ that is an AM radio broadcast signal is input to the synthesizer 12. Input signals S ₂ and S ₃ branched from the input signal S ₁ are input to the frequency transition device 20 and the transition frequency determination device 30, respectively.

The frequency transition device 20 includes a mixer 21 and a variable frequency oscillation device 22. The variable frequency oscillating device 22 generates a cosine wave having a transition frequency Δf _d determined by the transition frequency determining device 30 and a sine wave having a phase delayed by π / 2. To generate a cosine wave and sine wave, a real part and cosine wave, a sine wave as the imaginary part is to handle the frequency changes signal S ₄ as a complex number. Trends frequency determining unit 30, a self-seeking autocorrelation of the frequency characteristics of the AM radio broadcast wave signal output from the FFT unit 31, FFT unit 31 to fast Fourier transform of the input signal S ₃ is a AM radio broadcast signal (input signal) The correlation calculation device 32 and the Δf _d estimation device 33 for determining the transition frequency Δf _d from the maximum value of the autocorrelation obtained by the autocorrelation calculation device 32 are provided.

The variable frequency oscillating device 22 receives the transition frequency Δf _d determined by the transition frequency determination device 30 and inputs Acos (2πΔf _d t) and Asin (2πΔf _d t) to a real part mixer and an imaginary part mixer (drawing). The above is output to the mixer 21). The input signal S ₂ is input to the mixer 21, A cos output from the variable frequency oscillator _{22 (2πΔf d t), Asin} (2πΔf d t), and are multiplied respectively, the input signal S _2, the frequency Is output as a frequency transition signal S ₄ having a real part and an imaginary part that have shifted upward by Δf _d .

The frequency transition signal S ₄ is input to the equalization device 40 configured with a transversal filter. The equalizer 40 sequentially delays the input signal by the unit delay time τ, and a signal obtained by delaying the frequency transition signal S ₄ by the kτ delay time (k = 0 to Q−1) is output from each tap. . The unit delay time τ is equal to the sampling period Δt. Then, each signal delayed by each kτ delay time is multiplied by weight coefficients w ₀ ^{* to} w _Q-1 ^* , and the multiplied results are added. The equalizing device 40 is a device that corrects the frequency characteristics of the frequency transition signal S ₄ by weighting factors w ₀ ^{* to} w _Q-1 ^* and exp (j ωΔt) to exp (j ω (Q-1) Δt). is there.

等化装置４０は、伝達関数Ｚ（ω）のフィルタ特性により、周波数推移信号Ｓ₄ の周波数特性を修正した信号Ｓ₅ を、合成装置１２に出力する。合成装置１２は、ＡＭラジオ放送波である入力信号Ｓ₁ と、等化装置４０の出力信号Ｓ₅ とを合成して、合成信号Ｓ₆ を出力する。 The transfer function Z (ω) of the equalizer 40 is expressed by the following equation.

The equalizer 40 outputs a signal S _{5 in} which the frequency characteristic of the frequency transition signal S ₄ is corrected based on the filter characteristic of the transfer function Z (ω) to the synthesizer 12. The synthesizer 12 synthesizes the input signal S ₁ that is an AM radio broadcast wave and the output signal S ₅ of the equalizer 40 and outputs a synthesized signal S ₆ .

Further, the input signal S ₁ and the frequency transition signal S ₄ are input to the control device 50 that determines the weighting factors w ₀ ^{* to} w _Q-1 ^* by a PI (power inversion) algorithm. The control device 50, as the power of the combined signal S ₆ output from the synthesizer 12 is minimized, the weight coefficient of the equalizer ^{_{40 w 0 * ~w Q-1}} * is determined.

Next, the operation of the signal processing apparatus 1 will be described.
1. Operation of Transition Frequency Determining Device 30 The frequency characteristics of noise generated by a DC-DC converter having a basic frequency of 100 kHz basically have a large number of line spectra with a frequency interval of Δf, as shown in FIG. Δf is 100 kHz. The output of the FFT device 31 gives a frequency characteristic D (ω) that is a Fourier transform of the input signal S ₃ (t). Next, in the autocorrelation calculation device 32, the autocorrelation A (ω) of the frequency characteristic D (ω) is calculated by the following equation (2).

ＦＦＴ装置３１は、時間窓Ｔ_w におけるサンプリング点数ｕ＋１、サンプリング周期Δｔ＝Ｔ_w ／ｕのＦＦＴとすると、Δω＝２π／Ｔ_w の角周波数間隔で出力されるｕ＋１個の離散値となるので、離散化された自己相関Ａ（ｐΔω）は、次式の積和で演算される。

ただし、ｐ，ｑは、０，１，…ｕの整数変数である。また、ＦＦＴ装置３１の出力であるＤ（ｑΔω）は、周波数空間における１周期分のサンプリング値を表しているので、（３）式におけるＤ（ｑΔω）は、ＦＦＴ装置３１の出力単位（周期ｕΔω）を繰り返す周期関数として演算する。Ｄ（ｑΔω）の周波数帯域は、所望信号の帯域を含む帯域に設定する。Ｄ（ｑΔω）の周波数帯域を広くする程、雑音電力が大きくなるのでパワーインバージョンアルゴリズムを用いる場合には、所望信号の検出精度を高くすることができる。Ｄ（ｑΔω）の周波数帯域は、所望信号が精度良く検出でき、少なくとも所望信号の帯域を含む範囲にとる。

The FFT device 31 has u + 1 discrete values output at an angular frequency interval of Δω = 2π / T _w , assuming that the number of sampling points is u + 1 in the time window T _w and the sampling period is Δt = T _w / u. The discretized autocorrelation A (pΔω) is calculated by the product sum of the following equation.

However, p and q are integer variables of 0, 1,. Further, D (qΔω), which is the output of the FFT device 31, represents a sampling value for one period in the frequency space, so D (qΔω) in the equation (3) is an output unit (period uΔω) of the FFT device 31. ) Is repeated as a periodic function. The frequency band of D (qΔω) is set to a band that includes the band of the desired signal. As the frequency band of D (qΔω) is increased, the noise power increases. Therefore, when the power inversion algorithm is used, the detection accuracy of a desired signal can be increased. The frequency band of D (qΔω) is in a range in which the desired signal can be detected with high accuracy and includes at least the band of the desired signal.

As is clear from the spectrum of FIG. 2, the autocorrelation A (pΔω) increases at the frequency position of the spectrum where the switching noise of the DC-DC inverter is strong. That is, the autocorrelation A (pΔω) basically takes a local maximum value every angular frequency interval 2πΔf. Speaking of the integer variable p, the autocorrelation (pΔω) has a maximum value at an angular frequency where p is an integral multiple of 2πΔf / Δω. Therefore, each interval of periodicity (strict periodicity is not required) in the frequency characteristic of noise can be determined from the maximum value of autocorrelation. Of the frequencies at which autocorrelation A (pΔω) has a maximum value, any one of non-zero frequencies is selected as transition frequency Δf _d . The autocorrelation is maximized when there is no frequency transition, and is excluded in that case. In the present embodiment, as Δf _d , the frequency at which the maximum value of the autocorrelation A (pΔω) is maximized is selected. Basically, the frequency at which the maximum value is maximum is the smallest frequency Δf. Therefore, Δf _d = Δf is set.
Autocorrelation A (pΔω) is equal to the square frequency characteristic of input signal S ₁ on the time axis. Therefore, Δf _d may be obtained from the maximum value of the square frequency characteristic.

2. Operation of Equalizer 40 and Control Device 50 The control device 50 is a device that determines a weight coefficient in the equalizer 40. The control device 50 generates a reception vector x (pΔT) from the input signal S ₁ and the frequency transition signal S _{4 according} to the equation (4). Δt is a sampling period, and is equal to the unit delay time α in the equalizer 40. However, Δt and α (> Δt) are not necessarily equal. Q is the number of signals delayed by Δt, and is equal to the number of weighting factors of the equalizing apparatus 40 minus one. In this embodiment, the weighting coefficient is calculated at intervals of QΔt = ΔT. The input signal S ₁ (pΔT) means the p-th signal every ΔT period. The frequency transition signal input signal S ₄ (kΔt + pΔT) means the k-th signal for every Δt period in the pΔT period. However, k = 0, 1,..., Q-1. The received signal vector x (pΔT) is a Q + 1-dimensional column vector and is generated for each ΔT. ΔT is also a time interval at which the frequency transition signal input signal S ₄ performs a product-sum operation of weighting factors.

平均は、ＰΔＴ（＝ＰＱΔｔ）期間の平均であり、Ｐは平均する期間のΔＴ期間の数である。ただし、ｘ^H （ｐΔＴ）は、受信信号ベクトルｘ（ｐΔＴ）の複素共役の転置行列を表し、Ｑ＋１次元の行ベクトルである。したがって、平均相関行列Ｒは、Ｑ＋１次元の正方行列である。
図５に示すように、平均相関行列Ｒを演算する時刻を現在時刻とすると、過去ＰΔＴの期間において、ΔＴ期間毎に生成された受信ベクトルｘ（ｐΔＴ）を用いて、平均相関行列Ｒが演算される。すなわち、平均相関行列Ｒは、合成信号Ｓ₆ を求めるΔＴ期間毎に更新される。 Next, the time average R (hereinafter referred to as “average correlation matrix”) of the correlation matrix of the received signal vector is calculated by Equations (5) and (6).

The average is an average of PΔT (= PQΔt) periods, and P is the number of ΔT periods of the average period. However, x ^H (pΔT) represents a transposed matrix of complex conjugate of the received signal vector x (pΔT), and is a Q + 1-dimensional row vector. Therefore, the average correlation matrix R is a Q + 1-dimensional square matrix.
As shown in FIG. 5, when the time when the average correlation matrix R is calculated is the current time, the average correlation matrix R is calculated using the reception vector x (pΔT) generated every ΔT period in the past PΔT period. Is done. That is, the average correlation matrix R is updated every ΔT period for obtaining the combined signal S ₆ .

拘束ベクトルｃは、（８）式で定義されるＱ＋１次元の列ベクトルである。これは、入力信号Ｓ₁ の重み係数を固定したことに等しい。入力信号Ｓ₁ は、そのまま合成装置１２に入力しているので、入力信号Ｓ₁ に対する重み係数は１である。

（７）式で決定された重みベクトルＷの複素転置ベクトルＷ^H を用いて、（９）式により合成信号Ｓ₆ を生成することができる。ただし、Ｗ^H は、（１０）式に示すように、等化装置４０で用いられる重み係数１，ｗ₀ ^* 〜ｗ_Q-1 ^* を成分とするＱ＋１次元の行ベクトルである。重み係数１は、入力信号Ｓ₁ に対する重み係数である。ｋは、周期ΔＴ毎に、合成信号Ｓ₆ を求める時の期間番号であり、ｋΔＴが現在時刻を表す。

In the power inversion algorithm, the weight vector W (Q + 1-dimensional column vector) can be obtained from the inverse matrix of the correlation matrix R and the constraint vector c by the equation (7).

The constraint vector c is a Q + 1-dimensional column vector defined by equation (8). This is equivalent to fixing the weighting coefficients of the input signal S _1. Since the input signal S ₁ is input to the synthesizer 12 as it is, the weighting factor for the input signal S ₁ is 1.

By using the complex transposed vector W ^H of the weight vector W determined by the equation (7), the synthesized signal S ₆ can be generated by the equation (9). However, W ^H is a Q + 1-dimensional row vector whose components are the weighting factors 1, w ₀ ^{* to} w _Q-1 ^* used in the equalizer 40 as shown in the equation (10). Weighting factor 1 is a weighting factor for the input signal S _1. k is a period number when the composite signal S ₆ is obtained for each period ΔT, and kΔT represents the current time.

In this way, by determining the weighting factor and synthesizing the input signal S ₁ and the delayed synthesized signal S ₅ of the frequency transition signal S ₄ , the power of the synthesized signal S ₆ can be minimized. That is, when the noise power is larger than the power of the desired signal, the desired signal can be extracted by canceling the noise.

This is shown in FIGS. 2, AM radio broadcast signal DC-DC converter of a switching noise is superimposed, i.e., the frequency characteristic of the input signal S _1. When this frequency characteristic is shifted by the transition frequency Δf _d , the frequency characteristic shown in FIG. 3 is obtained. This frequency characteristic is a frequency characteristic of the frequency transition signal S _4. In the frequency characteristics of the input signal S ₁ and the frequency transition signal S ₄ , the positions of the switching noise spectra of the DC-DC converter overlap. As a result, to modify the frequency characteristics of the frequency transition signal S ₄ by the weighting factor ^{_{w 0 * ~w Q-1 *}} , when added to the frequency characteristic of the input signal S _1, erasing the switching noise of the DC-DC converter be able to. The frequency characteristic of the combined signal S ₆ at this time is as shown in FIG. At this time, two AM radio broadcast signals are generated: an original signal and a signal obtained by shifting the signal by Δf _d . However, since the transition frequency Δf _d is sufficiently wider than the frequency band of the AM radio broadcast signal of one station, at the time of channel selection in the AM radio receiver provided at the subsequent stage of the signal processing device for reducing noise in this embodiment, AM radio broadcast signals that have shifted by Δf _d can be removed.

  2, 3 and 4 are simulation results when radio power / DC-DC converter noise power = 0 dB and DC-DC converter noise power / receiver noise = 30 dB.

忘却係数αを大きくすれば、最新に求めた相関行列を平均相関行列Ｒに大きく反映させることができる。 In this embodiment, the average correlation matrix R is a simple average between the past PΔT, but may be as follows. Let R _{old be} the average correlation matrix between the past (P-1) ΔT, and let R _new be the correlation matrix in the latest ΔT for one period. A correlation matrix R for calculating the weight vector W may be obtained by equation (11).

If the forgetting factor α is increased, the latest correlation matrix can be greatly reflected in the average correlation matrix R.

ここで、ｓ^* （ｐΔＴ）は、既知の参照信号の列であり、拘束ベクトルｃは、Ｑ＋１次元の列ベクトルであり、参照信号と受信信号との相関ベクトルの平均である。このように拘束ベクトルｃを決定すれば、（７）式を用いた重みベクトルＷを求めることができる。このようにして重みベクトルＷから得られる（１０）式の重み係数１，ｗ₀ ^* 〜ｗ_Q-1 ^* を用いて周波数推移信号Ｓ₄ と入力信号Ｓ₁ とを合成すると、合成信号Ｓ₆ は、参照信号との誤差が最小となる。すなわち、参照信号ではない雑音は、除去されることになる。 In the first embodiment, the PI algorithm is used to calculate the weight vector W, but in the second embodiment, the least square error method is used. This example is a method of determining the constraint vector c of the first embodiment by the expression (12).

Here, s ^* (pΔT) is a known reference signal sequence, the constraint vector c is a Q + 1-dimensional column vector, and is an average of correlation vectors between the reference signal and the received signal. If the constraint vector c is determined in this way, the weight vector W using the equation (7) can be obtained. When the frequency transition signal S ₄ and the input signal S ₁ are synthesized using the weighting coefficients 1, w ₀ ^{* to} w _Q-1 ^* of the equation (10) obtained from the weight vector W in this way, the synthesized signal S ₆ The error from the reference signal is minimized. That is, noise that is not a reference signal is removed.

In the first and second embodiments, the equalization apparatus 40 is configured with a transversal filter. When the weight coefficient w _k ^* is obtained as in the first and second embodiments, the transfer function Z (ω) of the equalizer 40 is obtained from (1). Therefore, the frequency transition signal S ₄ (kΔt) is Fourier-transformed, the frequency characteristic FS ₄ (ω) is multiplied by the transfer function Z (ω), and the Fourier inverse transform is performed to obtain the delayed composite signal S ₅ (kΔt). The signal may be obtained and output to the synthesizer 12. Thus, the periodic noise can be canceled by synthesizing the input signal S ₁ (kΔt) and the delayed combined signal S ₅ (kΔt) on the time axis.

Examples 1-3 above, since realized by program processing of the CPU, the frequency transition device 20, (13) or as a device for multiplying the exponential function of the angular frequency 2Paiderutaf _d to the signal S ₂ by formula. It is clear that the input signal S ₂ shifts to the higher side by the frequency f _d .
In the _{first to} third embodiments, Δf _d is determined from the transition frequency based on the maximum value of the autocorrelation of the input signal S _1. However, the fundamental period is known as switching noise of the DC-DC converter. If the period does not vary with time, frequency transition may be performed using the known fundamental frequency of noise as transition frequency Δf _d instead of obtaining autocorrelation.

  In addition to the PI method and MMSE method described above, the algorithm for obtaining the weight vector W includes a generalized CMP method in which the constraint vector is set so as to reduce noise power, and a weight vector so as to maximize the S / N ratio. An MSN method for determining W can be used. The present invention combines a frequency transition signal obtained by shifting the frequency characteristics of an input signal by an integral multiple of the basic period of periodic noise and an input signal by determining a weighting factor so that noise power in the combined signal is reduced. It is the gist. Therefore, any algorithm can be used as long as it can be synthesized in such a manner. Further, the noise to be removed may not be completely periodic in the frequency space. That is, it is only necessary to have periodicity to such an extent that autocorrelation appears in the frequency space. Further, the present invention is applicable not only to the line spectrum but also to the continuous spectrum as the frequency characteristic of noise.

  The present invention can be used in an apparatus for removing periodic noise from an input signal.

12 ... Synthesizer 20 ... Frequency transition device 30 ... Transition frequency determination device 40 ... Equalizer

Claims (12)

In a signal processing apparatus for removing noise from an input signal in which noise is superimposed on a desired signal to be detected,
The noise has a frequency characteristic in which autocorrelation periodically varies with frequency,
A frequency transition means for generating a signal in which the input signal is frequency-shifted by an arbitrary frequency that is not zero among the frequencies at which the autocorrelation is maximized;
Equalizing means for correcting the frequency characteristics of the frequency transition signal output by the frequency transition means;
Synthesis means for synthesizing the signal output from the equalization means and the input signal;
And a control means for controlling a transfer function of the equalization means so as to suppress noise in the synthesized signal output from the synthesis means at least in a band including the desired signal.   2. The signal processing apparatus according to claim 1, wherein the frequency shifted by the frequency shifting unit is set to an integer multiple of a frequency determined by a reciprocal of a known fundamental period on the time axis of the noise. Obtaining the autocorrelation of the frequency characteristics of the input signal, obtaining a maximum value of the autocorrelation, and having transition frequency determining means for determining a non-zero frequency taking the maximum value,
The signal processing apparatus according to claim 1, wherein the frequency transition unit shifts the frequency of the input signal by a frequency determined by the transition frequency determination unit. The transition frequency determining means determines a non-zero frequency at which a maximum value of the autocorrelation is obtained, and a Fourier transform means for obtaining a frequency characteristic of the input signal; an autocorrelation of the frequency characteristic obtained by the Fourier transform means; Autocorrelation means;
The signal processing apparatus according to claim 3, comprising:   The signal processing apparatus according to claim 3, wherein the transition frequency determination unit obtains an autocorrelation of a frequency characteristic of the input signal from a square frequency characteristic of the input signal.   The signal processing apparatus according to claim 1, wherein the equalization unit is a unit that outputs a convolution of the impulse response of the transfer function and the frequency transition signal.   The equalizing means is a means for performing a transversal filter process in which the transfer function is realized by a sequential delay between taps and a weighting factor for multiplying a signal delayed and branched by each tap. The signal processing device according to any one of claims 1 to 6. The said equalization means is a means which calculates | requires the frequency characteristic of the said frequency transition signal, and outputs the impulse response of the product of the frequency characteristic and the said transfer function, The said any one of Claim 1 thru | or 5 characterized by the above-mentioned. 2. The signal processing device according to item 1.   9. The signal processing apparatus according to claim 1, wherein the control unit determines the transfer function by a power inversion algorithm that minimizes the power of the combined signal.   9. The signal processing apparatus according to claim 1, wherein the control unit determines the transfer function that minimizes an error between the synthesized signal and a known reference signal.   The signal processing according to any one of claims 1 to 10, wherein the desired signal is an AM radio broadcast wave, and the noise is noise generated by a DC-DC converter in a hybrid vehicle. apparatus. In a signal processing method for removing noise from an input signal in which noise is superimposed on a desired signal to be detected,
The noise has a frequency characteristic in which autocorrelation periodically varies with frequency,
A frequency transition signal in which the input signal is frequency-translated by an arbitrary frequency that is not zero among the frequencies at which the autocorrelation is maximized is generated,
Modify the frequency characteristics of the frequency transition signal,
Combining the modified frequency transition signal and the input signal;
A signal processing method characterized by correcting the frequency characteristics of the frequency transition signal so as to suppress noise in the synthesized signal at least in a band including the desired signal.

Images