JPS6326124A - Noise processing unit - Google Patents

Abstract

PURPOSE:To eliminate not only an impulsive noise but also an usual noise by applying real time DFT so as to calculate the spectrum of a signal in the frequency space, applying arithmetic processing thereto and applying IDFT to the signal so as to decode the signal into the original signal. CONSTITUTION:A voice signal (discriminator output) X (t) being an input enters a sample-and-hold circuit 26 through a low pass filter 24 preventing a reflection error, is sampled by a sampling frequency fs, digitized by an A/D converter 16 into a data x(n), which is subject to discrete Fourier transformation (DFT) to obtain a spectrum X(k). A component corresponding to a noise frequency component is excluded (nullified) from the X(k), inverse Fourier transformation (IDTF) is applied, the result is DA-converted and given to an interpolation LPF 24 to obtain a voice signal eliminating noise. Since a specific spectrum is eliminated completely, not only the impulsive noise but also the entire usual noise is eliminated.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a noise processor that produces an output from which noise contained in an audio signal has been removed, and is suitable for installation in a vehicle-mounted receiver or the like.

[Conventional technology]

In a vehicle-mounted receiver, noise is introduced into the audio signal due to interference caused by multiple reflected waves and direct waves arriving at the antenna, as well as impulses generated by the igniter and motor, so it is necessary to remove this noise.

A circuit shown in FIG. 11 or 12 is used to remove pulse noise in a conventional audio circuit. FIG. 11 shows a circuit that detects the start of pulse noise, holds the signal level at the level before the start of the pulse for a certain period of time, and then smoothes the signal waveform to remove the pulse noise. fa) is the pulse noise N
1 to N3, (bl is the audio signal after noise removal and smoothing, (C) is the gate pulse used for pulse noise removal. This method is based on the generation timing of the gate pulse and the pulse It is effective if the width is appropriate, but if the timing is off or the pulse width is inappropriate, the noise cannot be removed (Fig. (the noise could not be removed sufficiently), and instead added noise to the signal, resulting in an unnatural result.

In addition, as shown in FIG. 12, simply a low-pass filter LPF
There is also a method that smoothes the audio signal (
It is valid if high frequency noise is present in al, and (C
), a voice signal with the noise removed can be obtained, but
If the frequency component of the noise is close to the frequency band of the audio signal, it is difficult to remove it. In other words, an analog filter has a tail (so if you design an LPF to reproduce audio signals up to 10 KHz, noise with frequencies at the tail will pass through. If you lower the fc of the LPF to avoid this, the audio signal will be lowered). The treble part of the sound is cut off.

[Problem that the invention seeks to solve]

Recently, however, a method has been developed in which an audio signal is sampled, digitized, processed, and then converted back to analog to produce sound from a speaker.

By utilizing this digitization, a filter with ideally steep cutoff characteristics can be obtained, and the above-mentioned problems caused by drawing the tail of an analog filter can be eliminated.

The internal circuit of the receiver also includes elements with nonlinear characteristics, which cause output distortion. If a digital filter is used as the filter and its steep cutoff characteristic removes the high frequency components generated by the distortion, it is possible to reduce or improve the output distortion.

The present invention focuses on this point, and attempts to sample and digitize audio signals, and remove noise and nonlinear distortion using a digital filter.

[Means for solving problems]

The present invention provides a noise processor that is used in a receiver and outputs a signal from which noise contained in an audio signal has been removed. 26,16) and the first
A signal processing unit (30
), and a second circuit (
20, 22).

[Effect]

According to this noise processor, a specific spectrum can be completely removed, so it is possible to effectively remove not only impulsive noise represented by FM multipath distortion but also general noise in general. .

〔Example〕

In the present invention, the audio signal x (t) is sampled, each sample is digitized, N data x (n) is obtained, and this is subjected to discrete Fourier transform (DFT).
) to obtain the spectrum X(k). And this X (k
Of l, those corresponding to noise frequency components are removed (set to 0), then subjected to inverse discrete Fourier transform (IDFT), DA converted, and passed through an interpolation LPF to obtain a noise-free audio signal. This method is adopted.

These will be explained one by one below.

DFT of data x(n) is expressed by the following equation.

However, if O≦k<N-1, then...(3). After Σ on the right side of equation (3), x (nl g
(n) and h(n)
n), and equation (3) can be calculated using the circuit shown in FIG. 12.14 is a digital filter where h (nl) and Σ ((x(nln=.

g(n))h(k-n)). By performing the transformation according to equation (2) in this way, DFT can be performed using the chirp Z transform method, and k=0.1.2. ......, the spectrum will appear at the output end: X(0), X(1)
, X(21,...X(N-1)) appear one after another.

When you need all 8 X(g)) up to = 0-N-1, g(nl, g(k) is
n.

k is O to N-1), h(nl must be a sequence of 2N-1 points (n is -N+1 to N-1).

Details of FIG. 3 are shown in FIG. 4. Audio signal x (t) is A
/D converter 16 performs sampling and A/D conversion,
Output (n). FIG. 5 shows this sampling state. The sampling period is T, and the number of samples in one block is 1.
It is set at 28. Re(x (n)) in Figure 4 is x
(The real part of nl is 1m(x(nl) is the imaginary part,
Only the real part is obtained from the A/D converter. cos
πn/r'L −J stngn/N indicates g(n); g(n) = exp (j yr n and N) =
cosrcn2/N-j 5in1cn”/N)
, this is multiplied by the output of the A/D converter 16 by the multiplier ×,
If the adder Σ calculates the sum, x (nl g (n) is obtained. This is input to the filter 10. The filter output is multiplied by g(k), and the actual value of X(k) is obtained from the adder Σ. The part and imaginary part were output, and these were processed by the processor 18 (
After noise removal and IDFT), D/A converter 2
Enter 0 and return to analog.

FIG. 6 shows the real and imaginary parts of X(g)) for x(n) in FIG. Re(X(kl) is an even function with respect to N/2, and X(1)=X (127), x
+2>=x(126), . . . ■m'〔
X(g))] is an odd function and −X(1)=X (127
), -X(2)=X (126).

It becomes... The horizontal axis in FIG. 6 is k, which corresponds to frequency, and the interval between each value corresponds to 1/NT. Now, if the sampling frequency fs is 40KHz, then 1/NT=40/128 (KHz) and =
128 is N times that, so it corresponds to 40KHz,
N=64 in the middle corresponds to 20 KHz.

The processor 18 performs noise removal. This is spectrum X
This is done by setting the unnecessary portion of (k) (a portion corresponding to noise, for example, 10 KHz or more) to 0. To explain with reference to FIG. 7, it is assumed that noise N is added to the audio signal xft) as shown in the figure, and this is sampled and
．． Assume that 128 samples X Tnl are obtained each for n, III, . . . . Nikki DF for each block
T and calculate the real and imaginary parts of the spectrum Re(X(kl)
, T m (X(k)) is obtained. As shown in the figure, these are obtained sequentially with a delay of one block. Next, these R
Let the unnecessary part of e (X(k)) and l m (X(k)) be 0. These are symmetrical so you need to do them on both sides of the center. For example, if you want to cut frequencies above 10KHz, set X (32) to
32-96 are excluded.

Next, remove unnecessary parts Re (X(k)), I
Inverse discrete Fourier transform (IOFT) is applied to m[X(g)].
) to obtain the time domain signal y(n), which is converted to D/
A-converted and passed through an interpolation LPF to obtain a noise-free audio signal) F (t).

However, Oak≦N−1 Here kn=−(k +n −(n−k) ) −=(5
1, the equation (4) becomes. FIG. 8 shows how this is executed, and FIG. 9 shows the details. Other details conform to Figures 3 and 4.

FIG. 10 is a combination of FIG. 4 and FIG. 9, with the DFT section of FIG. 4 on the left and the IDFT section of FIG. 9 on the right. DFT
If you input X (n) into the IDFT section, y (n)
is obtained, and to=ixj (in the above example, to=32~
96), noise is removed from y(nl (however, the processor 18 is omitted in FIG. 10).

FIG. 1 shows a specific example of the present invention. This is the portion of the radio receiver shown as circuit 50 of the present invention in FIG. Second
In the figure, 40 is an antenna, 42 is a high frequency amplifier, 44 is a mixer, 46 is a local oscillator, 48 is an intermediate frequency amplifier, 52 is a discriminator, 54 is an audio amplifier, and 56 is a speaker, these are normal superhetero gain type. Configure the receiver.

In the present invention, the circuit 50 of the present invention is connected in parallel to the audio amplifier 54 by switches 58 and 60, and by operating these switches, the speaker 42 is driven by the output of the circuit of the present invention instead of the output of the audio amplifier. Make it.

As shown in FIG. 1, the circuit 50 of the present invention includes a low-pass filter 24, a sample-and-hold circuit 26, an analog-to-digital converter 16, a digital-to-analog converter 20. ? ! low-pass filter 22 for use in between, signal processing unit (processor) 30
The signal processing section consists of a memory (RAM) 32, an arithmetic section 34, a control section 36, and a memory (ROM) 38.

The input audio signal (discriminator output) x(t) passes through the low-pass filter 24 for preventing aliasing errors, enters the sample-and-hold circuit 26, is sampled at the sampling frequency fs, and is sent to the A/D converter 16 (discriminator output). In Figure 4, etc., this includes 26) and is digitized into data x (n). X(n) undergoes the above processing in the processor 30 to remove noise, then enters the D/A converter 20 to be converted into an analog signal, and is restored to an audio signal by the interpolation LPF. The circuit 50 of the present invention is opened with noise by operating the switches 58 and 60.
It is more effective in terms of sound quality etc. to operate the audio amplifier 54 and input its output to the speaker 56, and input the output of the audio amplifier 54 to the speaker when no noise is present.

The memory 38 of the processor 30 is used for storing processing programs, and the memory 32 is used for storing data. As mentioned above, N=
128, then 1281 x (nl, where n=0.1, 2°...) for each period I, II,...
. . 127 are sequentially stored in the memory 32, and DFT is performed on these data with a delay of one period, and IDFT is performed with a delay of one period. To list: ■A/D converter 1
6 output data x (nl import ■ previously imported 1
DFT for 28 data x in),
Calculation ■IDFT for the 128 previously calculated X fk)
.

y(n) is calculated, and (1) 7 (n) is converted to analog by the D/A converter 20. These steps (1) to (2) are repeatedly processed.

The frequency resolution Δf obtained when performing spectrum analysis is 1
/NT. >11 is symmetrical, so you only need to calculate the first half. The first spectrum X (O) comes out when the first N point data has entered the processor 30.

When the spectrum at N point frequencies from X(0), that is, frequency 0 to (N-1)Δf, is calculated, the frequency returns to frequency O again. As seen from the sampling theorem = N/2
=64, frequency fmax=64X1/NT=20KH
2 (assuming N=128.1/T=40KHz) is the highest frequency of this circuit.

〔Effect of the invention〕

As explained above, the present invention performs real-time DFT,
FM multipath Not only impulsive noise represented by distortion, but also normal noise can be effectively removed. This method is especially effective in cases where the frequency components of the noise are known, since they can be effectively removed, and are excellent for enjoying clear audio by removing harsh noises using car receivers.

[Brief explanation of the drawing]

1 and 2 are block diagrams showing an embodiment of the present invention, FIGS. 3 and 4 are block diagrams of the DFT section, FIGS. 5 to 7 are graphs explaining the DFT, and FIGS. Fig. 9 is a block diagram of the I DFT section, and Fig. 10 is a block diagram of the DFT section and the I DFT section.
The block diagram of the DFT section, FIGS. 11 and 12, are graphs explaining conventional noise removal procedures. In FIGS. 1 and 2, 24 is a low-pass filter, 26
1 is a sample hold circuit, 16 is an A/D converter, 20 is a D/A converter, 22 is a low-pass filter, 30 is a signal processing section, 40 is an antenna, 42 is a high frequency amplifier, 48 is an intermediate frequency amplifier, and 56 is a speaker It is. Applicant Fujitsu Ten Ltd. Representative Patent Attorney Minoru Aoyagi-CヱヱFigure 8t) Figure 9Figure 11Figure 12

Claims (1)

[Claims] In a noise processor that is used in a receiver and outputs a signal from which noise contained in an audio signal has been removed, a first noise processor that samples the audio signal, converts each sample into digital data, and sequentially outputs the digital data. circuit (26, 16), and signal processing that performs discrete Fourier transform on the output of the first circuit to obtain a spectrum, performs inverse discrete Fourier transform on this, and processes the spectrum to produce an output from which noise is removed. a second circuit (20,
22) A noise processor comprising: