JPH06209232A - Synchronous filter device - Google Patents

Abstract

PURPOSE:To reduce the effect of disturbance, and to improve the transient characteristic of a synchronous noise by correcting a gain so that the sum of squares of an error between the waveform of the inputted synchronous noise and the output waveform of a synchronous filter. CONSTITUTION:A pulse generator 11 divides the synchronization of the inputted synchronous noise by the number N of taps of a synchronous filter 12, and generates a sampling enabling pulse, and the filter 12 successively outputs the N pieces of taps synchronously with the pulse. The output of the filter 12 is added to the noise, and a filter coefficient correcting means 13 updates the tap value of the filter 12 so that the addition error becomes small. Next, the sum of squares of the difference between the output waveform of the filter and the waveform of the input noise is searched on the assumption that each tap of the filter 12 is multiplied by a gain (g) each time the pulse generated once in one cycle of the synchronous noise from the synchronous noise generating means 14 is inputted, and a filter gain correcting means 15 decides the gain (g) so that the sum of squares becomes minimum. Thus, the transient characteristic of the synchronous noise can be improved.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used for canceling noise synchronized with an engine, such as vibration or noise in the interior of an automobile, and eliminates the periodic noise in synchronization with the noise. The present invention relates to a progressive synchronous filter device.

[0002]

2. Description of the Related Art A synchronous filter device eliminates periodic noise in synchronization with its period, so that even if the period of noise changes, it adapts to it, and for example, a digital signal processing circuit (hereinafter , DSP), a normal FIR (Finite Respo)
(nse Filter) does not require a lot of sum-of-products calculation for filter calculation and can be configured very simply, so that it is widely used as a filter for canceling periodic noise.

A conventional synchronous filter device will be described below with reference to the drawings. FIG. 4 is a schematic block diagram of a conventional synchronous filter device. In FIG. 4, 41 is a pulse generator, 42 is a synchronous filter,
Reference numeral 43 is a filter coefficient correction means.

Next, the operation of the above conventional synchronous filter device will be described. In FIG. 4, a pulse generator 41
Is the synchronous filter 4 for the period of the input periodic noise.
It is divided by the tap number N of 2 and pulses are generated so that exactly N samplings can be performed within the period of the input noise. Even when the noise cycle fluctuates, the pulse cycle also fluctuates according to the fluctuation of the cycle. The synchronous filter 42 is a filter that sequentially outputs the tap values of the filter in synchronization with the pulse generated by the pulse generator 41. For example, if it is realized by a DSP, it is composed of a memory having a tap length only. After outputting to the end of the filter, it returns to the first tap of the filter and outputs.
The output of the synchronous filter 42 is added to the periodic noise, and the synchronous filter 42 is configured to reduce the addition error by the filter coefficient correction means 43 based on the addition error.
The value of the corresponding tap of is updated. For example, when the time is k and the number of filter taps is N, the position j of the corresponding tap coefficient can be expressed by the following equation (1).

J = k mod N (mod N is an integer value modulo N) (1) where the tap coefficient at time k is Wk and the addition error is e
(K), where u is a step size parameter corresponding to the gain, the tap coefficient Wk can be expressed by the following equation (2).

Wk (j) = Wk−N (j) + 2ue (k) (2)

By the above operation, it is possible to realize a synchronous filter device that cancels periodic noise. (For example, CAI Digital Signal Processing, Corona Publishing Co., Ltd., Hidefumi Obata, pp. 134-136, or Transactions of the Society of Instrument and Control Engineers, Vol. 19, No. 3, p. 34-
(See "Synchronous adaptive filter" on page 39)

[0008]

However, in the above configuration, when a disturbance such as random noise that disturbs the periodicity is added to the input and enters, the value of each coefficient of the synchronous filter is affected. And the error accumulates, so
The step size parameter, that is, the gain of one filter tap coefficient update cannot be set to a large value.
On the contrary, if the gain is suppressed, there is a problem that, for example, when the amplitude of the periodic noise changes, the gain is small, so that the filter response is slow and the convergence performance in the transient state deteriorates.

In view of the above problems, it is an object of the present invention to provide an excellent synchronous filter device capable of improving transient characteristics against periodic noise while suppressing malfunction due to disturbance as much as possible. .

[0010]

In order to solve the above problems, the synchronous filter device of the present invention pays attention to the periodicity of noise and considers the error between the waveform to be erased and the output waveform of the synchronous filter in that period. A means for changing the gain of the entire filter to a single rate so as to minimize the sum of squares is newly added.

[0011]

The present invention can realize a synchronous filter which is less affected by disturbance and has improved transient characteristics with respect to periodic noise.

[0012]

Embodiments of the present invention will be described below with reference to the drawings. 1 is a schematic block diagram of a synchronous filter device according to a first embodiment of the present invention. In FIG. 1, 11 is a pulse generator, 12 is a synchronous filter,
Reference numeral 13 is a filter coefficient correcting means, 14 is a periodic pulse generating means, and 15 is a filter gain correcting means.

Next, the operation of the synchronous filter device according to the first embodiment will be described. In FIG. 1, a pulse generator 11 divides the cycle of the input periodic noise by the number N of taps of the synchronous filter 12, and generates a pulse so that exactly N samplings can be performed within the cycle of the input noise. To do. Even when the noise cycle fluctuates, the pulse cycle also fluctuates according to the fluctuation of the cycle. The synchronous filter 12 synchronizes with the pulse generated by the pulse generator 11 and sequentially outputs the tap values of the synchronous filter 12. For example, if it is realized by a DSP, it is composed of a memory having a tap length only. When the synchronous filter 12 has been output to the end, it returns to the first tap and outputs. The output of the synchronous filter 12 is added to the periodic noise, and the value of the corresponding tap of the synchronous filter 12 is updated so that the addition error is reduced by the filter coefficient correction means 13 based on the addition error. Go. For example, when the time is k and the number of filter taps is N, the position j of the tap coefficient is determined as in the equation (1), and the time k
Let Wk be the tap coefficient, the addition error be e (k), and the step size parameter corresponding to the gain be u.

Next, the value of the periodic noise which is the input waveform and the value of the output of the synchronous filter 12 are stored for one cycle, and the periodic pulse generating means 14 generates the periodic noise once for each cycle. The specified algorithm is used every time a pulse is input, and one cycle of the square of the difference between the output waveform of the filter and the input noise waveform when it is assumed that N taps of the synchronous filter 12 are multiplied by a gain g For the sum of
The gain g is determined so that the sum becomes the minimum. The filter gain correction means 15 determines the gain g as described above, multiplies each tap of the synchronous filter 2 by one rate, and
Use a new tap coefficient.

As described above, according to the first embodiment,
Even if the amplitude of the input noise changes, as long as the periodic waveforms are similar, it is possible to respond steadily to the change in gain, although there is a delay of one period, so improve the transient characteristics of periodic noise. You can In addition, even when the step size parameter shown in the equation (2), that is, the gain for one filter tap coefficient update is set to a small value due to the case where non-periodic noise is included, the gain g remains the step parameter. Can be determined irrelevant, which has the advantage of increasing the convergence speed.

Next, the gain determination algorithm used in the first embodiment will be described. It is assumed that the input noise at the discrete time i is Ai and the output of the synchronous filter 12 is Bi. The synchronous filter 12 outputs N taps in one cycle, and the difference between the waveform formed by the filter and the input noise waveform is 2 when the tap is multiplied by the one-rate gain g.
If the gain g is determined so that the sum of one cycle of the power of two is taken and the average thereof is minimized, the following expression (3) is established.

[0017]

[Equation 1] If g that satisfies Expression (3) can be determined, the tap value of the synchronous filter 12 is updated for each cycle as in Expression (4) using g. Wi = Wi * g (i = 1, ..., N) (4) Here, if the left side of the equation (3) is replaced with f, the following equation (5) is obtained.

[0018]

[Equation 2] Then, when g that satisfies the expression (5), f is a minimum value and a minimum value with respect to g, and therefore the partial differential of f with respect to g is 0.
Then, the equation (6) is established. ∂f / ∂g = O (6) Next, substituting the equation (5) into the equation (6) and rearranging about g gives the equation (7).

[0019]

[Equation 3] Therefore, if g is determined according to the equation (7), for the sum of one cycle of the square of the difference between the waveform formed by the filter and the input noise waveform, assuming that the tap is multiplied by the one-rate gain g, Its value can be minimized.

If the difference between Ai and Bi is ei,
Expression (8) is established. Ai = Bi + ei (8) Substituting this into equation (7), equation (9) is obtained. Even if the input noise Ai cannot be measured and only Bi and ei can be measured, there is an advantage that the gain g can be determined by the equation (9).

[0021]

[Equation 4]

Next, a second embodiment of the present invention will be described with reference to FIG.
Will be explained. In FIG. 2, 21 is a pulse generator, 22 is a synchronous filter, 23 is a filter coefficient correction means, 24 is a periodic pulse generation means, 25 is a filter gain correction means, and 26 is a transfer function P and vibrates an electric signal. An actuator for converting into an output, 27 is a control compensation filter, 28 is an actuator simulated transfer function filter having a transfer function P, 29 is a vibration-electric signal converting means for converting mechanical vibration of a G sensor or the like into an electric signal, and 30 is a transfer. 3 is an actuator simulated transfer function filter having a function R.

Next, the operation of the second embodiment will be described. In FIG. 2, the actuator simulated transfer function filter 28 simulates the transfer characteristic P of the actuator 26, and the impulse response (transfer characteristic P) sampled and converted into the pulse sampling period of the pulse generator 21 is generated therein. To be done. The filter coefficient correction means 23 compensates the delay due to the actuator 26 with the output of the actuator simulated transfer function 28 to the addition error between the input noise obtained as the output of the vibration-electrical signal conversion means 29 and the output of the actuator 26, and the compensation thereof. The tap coefficient of the synchronous filter 22 is corrected by the output. Specifically, at time k and the number of filter taps N, the position j of the tap coefficient is determined by equation (1), the m-th tap coefficient at time k is Wk (m), the addition error is ek, and the actuator simulation is performed. The impulse response generated by the transfer function filter 28 is c
(M) (m = 1, ..., N), where u is the step size parameter corresponding to the gain, the update formula of the filter is expressed by the following formula (10). Wk + 1 (m) = Wk (m) + 2 * u * ek * c (j-m + 1) (m = 1, ..., N) (10) For example, an algorithm called Filtered-X has this content. Equivalent to. At this time, the control compensation filter 2
The transfer characteristic R of 7 may be simply 1.

Further, instead of performing the correction as described for the actuator simulated transfer function filter 28, the inverse function (R = 1 / R) of the actuator 26 is controlled by the control compensation filter 27.
A transfer function corresponding to P) may be configured (actually, in order to prevent oscillation, R = Y / P, Y is often configured as a transfer function for phase compensation), and in that case, filter coefficient correction For example, the means 23 may correct the synchronous filter 22 by the same formula as in the first embodiment.

The addition error between the periodic noise which is the input waveform and the output of the actuator 26 and the actuator 2
As the value output to 6, the value for the past one cycle is always stored. On the other hand, the impulse response characteristic of the actuator 26 is measured in advance by system identification or the like, and is realized as an FIR filter by the actuator simulated transfer function filter 30. Next, the actuator 2 for each sampling
The output after passing through the actuator 26 is estimated by obtaining the product sum of the past one cycle of the value output to 6 and the actuator simulated transfer function filter 30.

Here, each time a pulse generated once by the periodic pulse generating means 24 in the cycle of the periodic noise enters, the filter gain correcting means 25 uses, for example, the algorithm described below, and the filter gain correcting means 25 uses the algorithm described below. The gain g is determined and synchronized so that the sum of one cycle of the square of the difference between the output after passing through the actuator 26 and the input noise waveform, assuming that N taps are multiplied by the one-rate gain g, is minimized. Each tap of the expression filter 22 is multiplied by one to obtain a new tap coefficient.

The algorithm will be described below. The input noise at the discrete time i is Ai, the filter output is Bi, the difference between Ai and Bi is ei, and the input noise is U.
i, the addition error between the periodic noise and the output of the actuator 26 is stored for the past one cycle, and the value is Wi.
(I = 1, ..., N). On the other hand, the actuator 2
The impulse response characteristics of No. 6 are measured in advance by system identification, etc.
It is realized as an IR filter. If the impulse response is Vi (i = 1, ..., N), then S (W
The output Bi after passing the actuator 26 can be estimated by i * Vn-i + 1). Once Bi is obtained, g can be determined by equation (9).

Even if the input noise Ai cannot be measured and the output Bi of the actuator 26 cannot be measured and only the addition error and the value to be output to the actuator 26 are known, the actuator 26 is estimated by estimating the output. The gain g can be determined so that the sum or average of the square of the difference between the output after passing and the input noise waveform is one cycle or the average thereof is minimized.

As described above, according to the second embodiment,
Even if the amplitude of the input noise changes, as long as the periodic waveforms are similar, it is possible to respond steadily to the change in gain, although there is a delay of one period, so improve the transient characteristics of periodic noise. You can In addition, even if the step size parameter u in the equations (2) and (10), that is, the gain for one filter tap coefficient update is set to a small value in the case where non-periodic noise is included, the gain is stepped. Since it can be determined independently of the parameters,
There is also an advantage that the convergence speed can be increased.

Further, even when the actuator output added to the input noise is not directly obtained, the gain g can be determined by estimating the actuator output.

In the second embodiment, the sum for one cycle is taken, but it may be an integral multiple of the cycle, or the sum of time long enough to neglect the effect of the cycle or the average value is small. Even if it is done, the same good operation can be achieved.

Next, a third embodiment of the present invention will be described with reference to FIG.
Will be explained. In FIG. 3, 31 is a pulse generator, 32 _{1 to} 32 _M are synchronous filters, and 33 _{1 to} 33 _M.
Is a filter coefficient correcting means, 34 is a periodic pulse generating means,
Reference numeral 35 is a filter gain correction means.

Next, the operation of the third embodiment will be described. The present embodiment is an example in which M synchronous filters and filter coefficient correcting means in the first embodiment are respectively provided and the number of input noises to be erased is increased to P, and the basic operation is As already described in the description of the first embodiment, since the M synchronous filters cancel the P input noises, the synchronous filters 33 _{1 to} 33 33 are used.
_M gain Gj (j =, ..., M ) is the filter gain correction unit 35, synchronous filter j (j
= ,. ． ． , M) each tap is multiplied by the amplification factor Gj, and the sum of squares of the difference between the filter output and the waveform of the input noise is taken at P points, and the sum of the P sums of squares becomes small. Is decided.

Here, the algorithm for determining the gain used by the filter gain correction means 35 will be described. Let _{k k} (i) be the k-th input noise at discrete time i, and B _jk (i) be the component that eliminates the k-th input noise by the j-th synchronous filter 32 _j . 1 synchronous filter 32 _j between the period and outputs N taps, the output waveform and the input noise waveform of the filter when it is assumed that multiplied by the Ichiritsu gain g _j to the tap of the j-th synchronous filter 32 _j Difference 2
Taking the sum of one cycle of the power and determining the gain g so that the sum becomes the minimum, the following equation (11) is established.

[0035]

[Equation 5] If g that satisfies Expression (11) can be determined, then g
Is used to update the tap value of the j-th synchronous filter 32 _j as shown in Expression (12) every cycle. W _j (i) = W _j (i) * g _j (i = 1, ..., N) (12) Here, if the left side of equation (12) is replaced with f, the following equation (13) )become that way.

[0036]

[Equation 6]And now, the a-th synchronous filter 32_aGain g
_aAsk for. f is g_a( _a= 1 ,. ． ． , M) poles
Since it is a small value and the minimum value, g of f_aThe partial derivative with respect to
It becomes 0, and the equation (14) holds. ∂f / ∂g_a= 0 (14) Next, substituting equation (13) into equation (12) yields equation (15)
Like

[0037]

[Equation 7] Here, when each A _k (i) is approximated by the following equation,

[0038]

[Equation 8] Substituting equation (16) into equation (15) yields equation (1
It becomes like 7).

[0039]

[Equation 9] Here, when g _k other than g _a is close to 1 or has a constant value, g _a is obtained by the following equation (18).

[0040]

[Equation 10] Therefore, if g _a is determined according to the equation (18), the difference between the output waveform of the synchronous filter 32 _a and the input noise waveform can be calculated by multiplying the tap of the a-th synchronous filter 2 by the one-rate gain g _a . It is possible to minimize the sum of one cycle of the square. In addition, when g _k other than g _a greatly deviates from 1, it may be standardized as in the following equation (19). g _k = 1 + (g _k −1) / M (k = 1, ..., M) (19)

[0041]

As described above, according to the present invention, even if the gain of the step parameter of the synchronous filter is reduced so that the influence of disturbance is reduced as much as possible, it is possible to quickly respond to the gain fluctuation of the periodic noise, and It has an effect that the characteristics can be improved.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of a synchronous filter device according to a first embodiment of the present invention.

FIG. 2 is a schematic block diagram of a synchronous filter device according to a second embodiment of the present invention.

FIG. 3 is a schematic block diagram of a synchronous filter device according to a third embodiment of the present invention.

FIG. 4 is a schematic block diagram of a conventional synchronous filter device.

[Explanation of symbols]

 11, 21, 31 Pulse generator 12, 22, 32 Synchronous filter 13, 23, 33 Filter coefficient correction means 14, 24, 34 Periodic pulse generation means 15, 25, 35 Filter gain correction means 26 Actuator 27 Control compensation filter 28 Actuator simulated transfer function filter 29 Vibration-electric signal conversion means 30 Actuator simulated transfer function filter

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Takeshi Okada 4-3-1, Tsunashima-higashi, Kohoku-ku, Yokohama, Kanagawa Matsushita Communication Industrial Co., Ltd.

Claims (6)

[Claims] 1. A pulse generator that generates a pulse that can be sampled by dividing a period of a periodic component of input noise into N, and a periodicity of the input noise that sequentially outputs N taps in synchronization with the pulse. A synchronous filter that eliminates components,
Two of the difference between the filter output and the waveform of the input noise, assuming that each tap of the filter is multiplied by the amplification factor G
A synchronous filter device comprising a filter gain correction means for taking a sum of multiplications and determining an amplification factor of each filter so that the sum of squares becomes small. 2. The discrete time i, the input noise Ui, and the filter output Wi are the sum of one cycle or n cycles as S.
When expressed by (), S (Ui * Wi) / S (Wi
The synchronous filter device according to claim 1, wherein * Wi) is the amplification factor of the filter. 3. A discrete time i, an input noise Ui, a filter output Wi, and a difference between the input noise Ui and the filter output Wi is Ei, and a sum of one period or n periods is S ().
When expressed by, S (Ei * Wi) / S (Wi * W
2. The synchronous filter device according to claim 1, wherein i) +1 is the amplification factor of the filter. 4. A pulse generator that generates a sampleable pulse by dividing a period of a periodic component of input noise into N, and sequentially outputs N taps in synchronization with the pulse,
A synchronous filter that eliminates the periodic component of the input noise after passing the elements of the transfer characteristic P, and each tap of the filter is multiplied by an amplification factor G, and the output after passing the filter is used to transfer the transfer characteristic. A simulated transfer function filter for estimating the output Vi 'after passing through P and the sum of squares of the difference between the output Vi' and the periodic noise are taken, and the amplification factor of each filter is reduced so that the sum of squares becomes small. And a filter gain correction means for determining 5. At a discrete time i, the output of the tap value of the filter in the past m hours or the output of the control compensation filter in the past m hours is set to Q = (Qm, Qm-1, ..., Q1), Further, a characteristic that is the same as or close to the transfer characteristic P is FIR filter P '= (P1, P
2 ,. ． , Pm−1), the output Vi ′ after P is estimated by the product sum v of Q and P ′, and from the difference E ′ between the input Vi ′ and the input noise Ui and the true P output Vi. , When one cycle or the sum of n cycles is represented by S (), S
5. The synchronous filter device according to claim 4, wherein ((Ei * v'i) / (V'i * V'i) +1) is the amplification factor of the filter. 6. A pulse generator that generates a sampleable pulse by dividing a period of a periodic component common to P input noises into N, and outputs N taps in sequence in synchronization with the pulse, M synchronous filters for eliminating the periodic components of P input noises, and the filter j (j
= ,. ． ． , M) each tap is multiplied by the amplification factor Gj, the sum of squares of the difference between the filter output and the waveform of the input noise is taken at P points, and the sum of the P sums of squares becomes small. Thus, the amplification factor Gj (j
= ,. ． ． , M) for determining the filter gain correction means.