JPH0677768A - Sampling rate converting method for digital audio data - Google Patents

Abstract

PURPOSE:To directly transform digital audio data, which are sampled at a certain sampling rate, to digital audio data at a different sampling rate. CONSTITUTION:The sampling rate of the source digital audio data is defined as f1 and the sampling rate of the converted digital audio data is defined as f2. The integer part of f1/f2 is defined as N, the remainder is defined as tau%, and an i-th sampled value Di of the f2 after the coincidence of both sample points is calculated as the sum of respective products of an N-th sampled value and ten sampled values before and after it, of the source data, and transformation coefficients. The transformation coefficients are calculated based on a function obtained by Fourier transforming a single rectangular pulse and further inversely Fourier transforming the result by using a high frequency cut filter.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for converting digital audio data sampled at a predetermined sampling rate into digital audio data sampled at another sampling rate.

[0002]

2. Description of the Related Art In recent audio equipment, audio data is often treated as digital data from the viewpoint of improving S / N, reducing crosstalk and distortion. Moreover, gain control and mixing have come to be performed digitally.
However, as the sampling rate of digital audio data, various sampling rates have been adopted with the development of digitization. For example, the audio file of AM broadcasting station adopts a sampling rate of 24KHz, the audio transmission bit rate for FM broadcasting station is 32KHz, and it is 4K for audio data recording using VTR.
The sampling rate of 4.065KHz is adopted, and the sampling rate of digital audio data of CD is 44.1.
KHz is adopted, and 48KHz is adopted as the bit rate of digital audio data in the FM broadcasting station.

[0003]

As described above, the various digital audio devices currently in use have different sampling rates. Therefore, when the output signals of these devices are mixed, for example, the sampling rates are unified. It has to be one. That is, it is necessary to convert the sampling rate of digital audio data.

As a simple method for converting the sampling rate of digital audio data, it is conceivable to convert all digital audio data into an analog signal once and then resample this analog signal at a desired sampling rate. However, once the digital audio data is converted into an analog signal,
In that case, good S / N, crosstalk, distortion rate, etc., which are characteristics of digital audio data, are lost.

An object of the present invention is to eliminate the above-mentioned conventional drawbacks and to convert the sampling rate of digital audio data to another sampling rate without converting the digital audio data into an analog signal. An object of the present invention is to provide a sampling rate conversion method for digital audio data in which original digital audio data and digital audio data whose sampling rate has been converted correspond to each other with a high degree of accuracy.

[0006]

According to the present invention, in converting the sampling rate of digital audio data, when the original sampling rate of digital data is f ₁ and the new sampling rate is f ₂ , both sampling rates are changed. From the moment when the sampling timings match, the i-th sampling timing of f ₂ is set to f ₁ · i
/ f ₂ = N + τ, where N is the integer part and τ is f ₁ · i
/ f ₂ is the fractional part of the remainder expressed as a percentage, and is the conversion when calculating the i-th sample value from the N-th sample value of the original digital audio data and about 10 sample values before and after it. The coefficients are determined based on the function obtained by Fourier transforming a single rectangular pulse and then inverse Fourier transform with a high pass cut filter, and these transform coefficients are pre-determined for all possible percentages. A sample value of a sampling rate, which is stored in a memory and is converted by performing a product-sum operation of the conversion coefficient read from the memory and the N-th sample value and the sample values before and after the sample value. Is.

[0007]

According to the sampling rate conversion method for digital audio data of the present invention, the sampling rate can be converted to the desired one without converting the original digital audio data into an analog signal. Therefore, the sampling rate can be converted without impairing the excellent characteristics originally possessed by the digital audio data. Furthermore, since the transform coefficient is determined based on the function obtained by Fourier transforming a single rectangular pulse and then inverse Fourier transform using a high-pass cut filter, it is possible to interpolate sample values with high fidelity. Therefore, the original digital audio data and the sampling rate-converted digital audio data are correlated with extremely high accuracy.

[0008]

1 is a block diagram showing the overall construction of an apparatus for implementing a sampling rate conversion method for digital audio data according to the present invention. In this example, it is assumed that digital audio data with sampling rates of 24 KHz, 32 KHz, 44.1 KHz and 48 KHz is handled. In the present invention, the sampling rate is converted in units of the period from the coincidence of sampling timing to the coincidence of sampling timing at a certain instant, so that the sampling timing of all digital audio data is synchronized. Need to let. For this purpose, the reference sampling pulse generator 11 is provided with one reference oscillator 12, and the reference signal generated from this reference oscillator is divided to generate the reference sampling pulse serving as the reference of the sampling pulse of each digital audio data. To do so.

The above-mentioned sampling rate of digital audio data is decomposed into prime numbers as follows.
^{24KHz = 2 6 3 1 5 3} , 32KHz = 2 8 5 3, 44.1KHz = 2 2 3 2 5 2 7 2, 48KHz =
2 ⁷ 3 ¹ 5 ³ . If these least common multiple (LCM) oscillators are used, all the above sampling rates can be obtained by dividing the output signal. However, in this example, the reference oscillator 12 is configured with an LCM × 2 oscillator so that the duty cycle of all sampling pulses is 50%. That is, the oscillation frequency of this reference oscillator 12 is 2 ⁸ 3 ² 5 ³ 7 ² × 2 = 28.224 MHz. The reference signals generated from this reference oscillator 12 are 1/294 and 1/32, respectively.
The frequency is divided by 0, 1/441, 1/588 dividers 13 to 16, and the output signals of these dividers are further divided by 1/2 by the flip-flops 17 to 20 to reduce the frequency as described above. 24KHz, 32K
Hz, 44.1KHz and 48KHz, 50% duty cycle
It is possible to obtain four types of reference sampling pulses synchronized with each other.

These sampling pulses are applied to each audio device.
Supply to 21-24. Each of these audio devices is provided with its own voltage-controlled oscillator 25-28, and its output pulse is appropriately frequency-divided by frequency dividers 29-32 to generate desired sampling pulses, respectively. Sampling pulses generated from these dividers 29-32,
The phase of the reference sampling pulse generated by the reference sampling pulse generator 11 as described above is compared by the phase comparators 33 to 36, and the PLL for controlling the oscillation frequency of the voltage controlled oscillators 25 to 28 is provided by the difference. Thus, a sampling pulse synchronized with the reference sampling pulse can be generated. Digital audio data can be obtained by sampling the audio signal from the audio data source in each of the audio devices 21 to 24 by the sampling pulse thus generated.

In this way, the digital audio data output from each of the audio devices 21 to 24 is supplied to the arithmetic circuit 37 which performs an arithmetic operation for sampling rate conversion. The arithmetic circuit 37 selects at least two out of the four input digital audio data, and makes their sampling rates coincide with each other. Now, for convenience of explanation, 48KH
The mixing shall be performed using digital audio data having a sampling rate of z and digital audio data having a sampling rate of 44.1 KHz. In this case, the 48 KHz digital audio data is used as it is, and the 44.1 KHz digital audio data is converted to the 48 KHz digital audio data.

As shown in FIG. 2, a certain sampling timing of 44.1 KHz digital audio data and 48
It is considered that a certain sampling timing of KHz digital audio data coincides with each other at time t ₀ . The sampling points of the 48 KHz digital audio data will arrive little earlier than the sampling points of the 44.1 KHz digital audio data. In other words, if one sample section of 44.1KHz digital audio data is equally divided into 480 sections, the sampling point of 48KHz digital audio data will be the 441th section. Therefore, in general, the i-th sample point counted from the time t ₀ of 48 KHz digital audio data is

[Equation 1] Will be given in. That is, the i-th sample point of 48KHz digital audio data is 44.1KHz.
It is shown that it is located between the N-th sample point and the N + 1-th sample point of the digital audio data of, and is located at a time separated by τ% from the N-th sample point.

Now, consider a process in which a digital signal is once converted into an analog signal by a D / A converter and the analog signal is sampled at a new sampling rate. Converting a digital signal to an analog signal,
The staircase waveform is as shown in. Quantization noise is removed from such a staircase waveform, and 1 /
An analog signal having a smooth waveform can be obtained by passing it through a high-frequency cut filter that cuts two or more frequency components.

Although the above-mentioned staircase waveform has different levels depending on each step, it can be considered as a continuous single rectangular wave including positive and negative signs. Therefore, it can be considered that the effect on the i-th sample point is centered on the N-th rectangular wave calculated earlier and the effects of the rectangular waves on several sample points before and after that are aggregated. The reason is that the effect of a single pulse from its center to its periphery is known to converge to zero fairly quickly. Now Figure 4A
The Fourier transform of a square wave with amplitude A and pulse width T as shown in

[Equation 2] And the waveform is as shown in FIG. 4B, the Fourier inverse transform cut at a frequency half the sampling rate is

[Equation 3] Therefore, as shown in FIG. 4C, a reproduced waveform of one rectangular wave can be obtained.

Therefore, in the present invention, -T / 2≤t <
Let F ₀ (τ) be the value for T / 2, and create a table showing the details in τ%. Similarly, −21T / 2 to −19T / 2,
The value for −19T / 2 to −17T / 2, ----− 3T / 2 to −T / 2 is F
_-10 (τ), F _-9 (τ), --- F _-1 (τ), T / 2 〜 3T / 2,
The value for 3T / 2 to 5T / 2, ---- 19T / 2 to 21T / 2 is F ₁ (τ),
Assuming that F ₂ (τ), --- F ₁₀ (τ), a table as shown in FIG. 5 is created. This table shows the conversion coefficient when τ is set. As described above, after the sample points match at a certain time t ₀ , the i-th sample value at the new sampling rate calculates the integer part N and the decimal part τ based on Equation 1, The Nth sample value of KHz, that is, the amplitude is A ₀ , the N-10th sample value is A _-10 , the N-9th sample value is A _-9, and similar sample values are A _-8 , A _-7 , --- A ₁ , A ₂ , A ₃ , --- A
Assuming ₁₀ , the i-th sample value B _{i at} 48 KHz is

[Equation 4] Can be asked as

As described above, in order to convert 44.1 KHz digital audio data into 48 KHz digital audio data, it is sufficient to convert 441 44.1 KHz data into 480 data, as shown in FIG. In 4
461 including 4.1KHz data 441 data d _{0 to} d ₄₄₀ and 10 data before and after it d _{-10 to} d _-1, d _{441 to} d ₄₅₀
A buffer memory capable of storing individual data is provided. In this case, the sample position of certain data converted to 48KHz is shown in Fig. 6.
The data may be the same as the data d ₀ . Assuming that some data converted to 48 KHz is D ₀ , as described above, 48
The i-th data D _i of KHz comes to the position deviated from the N-th data of 44.1 KHz by τ% from the equation (1). At this time, the value of this data D _i is obtained from the equation (4).

FIG. 7 shows an arithmetic circuit 37 for performing such an arithmetic operation.
An example is shown. Control writing to buffer memory 41
The write address circuit 42 to be controlled has a 44.1 kHz sampling
The read address that supplies the pulse
The sampling circuit 43 is instructed to sample data before and after the data group.
Supply the signal j that determines the address and the address of the N-10th data
The write address circuit 42 and the read address.
Buffer memory by switching output of switch circuit 43 with switch 44
41 so that it can be supplied to. In addition, the above calculation
A ROM 45 that stores the coefficient in advance is provided, and the coefficient value F_-Ten(τ), F
_-9(τ), --- F_-1(τ), F₀(τ), F₁(τ), F₂(τ),-
-F_TenRemember (τ). These coefficient values and
Data values stored in the buffer memory 41 are multiplied by the multipliers 46-1 to 46-2.
Feed to 1, find these products, and add these products to the adder 47
And perform the calculation according to the above-described equation (4). This addition
The value supplied from the calculator 47 is the sample clock of 48KHz.
48KHz sample by writing to register 48 at
Great digital audio data D_iIs obtained
Will be. This calculation is based on a sampling rate of 48 KHz.
Must be done within about 20 μs, which is one cycle of
Since it does not exist, it is necessary to use a high-speed arithmetic element.

Since the sampling conversion needs to be performed one after another, it is necessary to perform new data calculation again after 480 new data are obtained. Therefore, as the buffer memory 41, it is necessary to use a ring-shaped buffer memory as shown in FIG. That is, the data of 44.1 KHz is first written in the memory from the point A, and becomes 461 at the point B. At this point, centering on the 11th point X from point A
Conversion to 48KHz starts. 20 times per operation
It ends in μs and is taken into the register 48 by the clock of 48KHz. When the calculation progresses sequentially and 479 calculations are completed from point X, that is, when 480 data are output, 44.1
Since there is a fixed relationship between KHz and 48KHz, it means that data is written up to point D. Therefore, point Y is replaced by point X.
If you start the calculation with this, the next 480 pieces of data will be calculated, and at the 480th time you will return to the initial state. In this way, 44.1 KHz digital audio data can be continuously converted to 48 KHz digital audio data.

The above description has dealt with the case of converting the digital audio data of 44.1KHz to digital audio data of 48KHz, typically sampled between f ₁ and f ₂ in the case of converting the sampling rate from f ₁ to f ₂ The formula for the relative position of is

[Equation 5] Will be given in.

When converting f ₁ to f _{2 in} this way, f ₁ >
If it is f ₂ , since the sample data of f ₁ contains high frequency components, area sync may occur. At this time, preferably used is obtained by limiting the band of the inverse Fourier transform to _{f 2/2 F j (τ} ). 24KHz, 32KHz,
In the case of 44.1KHz and 48KHz, there are six cases of conversion to a high frequency, but it is sufficient to increase the memory capacity of the conversion coefficient by 8 times.

[0021]

As described above, according to the present invention, digital audio data having different sampling rates can be directly converted into digital audio data having a desired sampling rate without being converted into analog signals. S / when converting to
N, crosstalk, distortion, etc. are not deteriorated. In the conversion, a single rectangular pulse is Fourier-transformed, then the transform coefficient obtained by the inverse Fourier transform using a high-frequency cut filter is obtained, and the calculation is performed using about 10 data before and after. Very accurate conversion can be performed. As a result, mixing processing using a multiplier can be easily performed, which greatly contributes to digital processing of voice information.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an example of an apparatus that implements a sampling rate conversion method for digital audio data according to the present invention.

FIG. 2 is a diagram for explaining the principle of the conversion method according to the present invention.

FIG. 3 is a diagram showing a digital audio data D / A.
It is a graph which shows a staircase waveform at the time of conversion.

4A, B and C are graphs showing the Fourier transform and inverse Fourier transform of a single square wave pulse.

FIG. 5 is a diagram showing a conversion coefficient table.

FIG. 6 is a diagram showing a configuration of a buffer memory.

FIG. 7 is a block diagram showing a configuration of an example of an arithmetic circuit.

FIG. 8 is a diagram showing a configuration of a ring-shaped buffer memory.

[Explanation of symbols]

 11 Reference sampling pulse generator 12 Reference oscillator 13-16 Frequency divider 17-20 Flip-flop 21-24 Audio equipment 25-28 Voltage controlled oscillator 29-32 Frequency divider 33-36 Phase comparator 37 Arithmetic circuit 41 Buffer memory 42 Write address circuit 43 Read address circuit 44 Switch 45 ROM 46-1 to 46-21 Multiplier 47 Adder 48 Register

Claims (1)

[Claims] 1. When converting the sampling rate of digital audio data, when the original sampling rate of digital data is f ₁ and the new sampling rate is f ₂ , from the moment when the sampling timings of both sampling rates match ₂ of the i-th sampling timing, given by _{f 1 · i / f 2 =} N + τ, here
N is an integer part and τ is a decimal part that represents the remainder of f ₁ · i / f ₂ as a percentage. The i-th sample value is the N-th sample value of the original digital audio data and the values before and after it.
The transform coefficients when calculating from about 10 sample values are determined based on the function obtained by Fourier transforming a single rectangular pulse and then inverse Fourier transform using a high-pass cut filter. The values obtained in advance for all possible percentages are stored in a memory, and the product of the conversion coefficient read from this memory and the N-th sample value and the sample values before and after it are converted to perform conversion. A sampling rate conversion method for digital audio data, characterized by obtaining a sampled value of a sampling rate.