JP3458942B2 - Digital signal decoding method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for decoding a digital signal for decoding an encoded digital signal, and more particularly, it includes a step of scaling an operand value in an arithmetic operation for decoding. The present invention relates to a method for decoding a digital signal.

[0002]

2. Description of the Related Art Conventionally, a digital signal processing device (DSP; digital signal
A block floating technique is known as an encoding technique for rocessor. In this technique, a digital signal is divided into blocks for a predetermined time or a predetermined number of words, and floating processing is performed in block units.

An encoding technique called orthogonal transform encoding is also known in which an inputted digital signal on the time axis is subjected to orthogonal transformation, converted into spectrum data on the frequency axis and encoded. There is.

Further, before performing this orthogonal transform coding,
A technique is also known in which the block floating technique described above is applied to a digital signal to reduce the number of bits of the digital signal that is orthogonally transformed. And such a technique is disclosed in, for example, Japanese Patent Application Laid-Open No. 4-30254
No. 0 publication.

In the structure described in this publication, the block length in the block floating performed at the time of orthogonal transformation is variable. Then, in this configuration, in order to reduce the amount of calculation for determining the block length, the block length and the block floating coefficient are determined based on the same index of the digital signal, for example, the maximum absolute value of the signal. It has become so.

Further, Japanese Patent Laid-Open No. 6-164414 discloses a configuration for reducing a calculation error caused by a decrease in the number of bits of a digital signal. That is,
With this configuration, the scale-down is performed when the orthogonal transform calculation or the inverse orthogonal transform calculation is performed on the digital signal. And in this configuration,
The input data value in each calculation is compared with a preset scale-down judgment reference value, and the presence or absence of scale-down in each calculation process is judged.

[0007] Further, the inventors of the present application, when determining scale down, instead of the input data of each calculation formula, a representative value of the spectrum data of each unit (for example, a scale factor value in ATRAC voice encoding and decoding). It is proposed in Japanese Patent Application No. 9-300002 that the calculation amount required for the determination is reduced and the scale-down width is realized with a finer precision than a simple bit shift by using.

[0008]

However, in the configuration described in Japanese Patent Laid-Open No. 6-164414 mentioned above, since the scale-up or scale-down is performed based on the input data string of each operation process, the input signal is Although a signal of medium to small amplitude has some effect, it consumes a large amount of power and is not suitable for a small portable device.

Further, in the configuration described in Japanese Patent Laid-Open No. 4-302540, when the input signal has a large amplitude, the amount of calculation is smaller than that in the configuration described in Japanese Patent Laid-Open No. 6-164414 and the same precision calculation is performed. The result can be derived. However, when the input signal has a small amplitude, the calculation accuracy becomes poor.

Furthermore, in these conventional configurations, the effect of scale-up and scale-down is small when the input signal is a non-signal or a minute signal. Therefore, it is not possible to expect an improvement in calculation accuracy commensurate with an increase in the amount of calculation for scale-up or scale-down.

The present invention has been made to solve the above-mentioned conventional problems, and an object thereof is to provide a digital signal decoding method for performing highly accurate decoding without greatly increasing the amount of calculation. Is to provide.

[0012]

In order to achieve the above object, the method for decoding a digital signal according to claim 1 of the present invention is divided into a plurality of units according to a time range and a frequency domain, and each unit is divided into a plurality of units. In a digital signal decoding method for decoding spectrum data coded for each unit based on an index value set for each unit, a first value is searched from index values in each unit for a representative value. A step, a second step of determining whether or not scaling is performed on the operand value in the operation for decoding based on the representative value, and the operand value based on the determination result in the second step. And a third step of scaling.

In the above method, the spectrum data means MDCT (modified discrete cosine transform) for a digital signal obtained by sampling.
It is a digital signal in the frequency space obtained by performing processing and the like. Then, this spectrum data is divided into a plurality of units according to a predetermined time range and a predetermined frequency region. Here, the time range is, for example, a range of time when the digital signal is sampled.

A predetermined number of spectrum data belong to each unit, and the spectrum data is encoded for each unit. Then, the quantization for each unit is performed based on the index value in each unit. As the index value, for example, the maximum value in the amplitude value of the spectrum data belonging to each unit is used as described in claim 5.

In the above method, when decoding the spectrum data divided into such units,
In order to avoid overflow in the arithmetic processing in the decoding process, or to accurately decode data with a small amplitude, it is possible to perform scaling on the operated value. The calculated value is an input value in each calculation process, that is, data related to the calculation.
Scaling is to enlarge (scale up) or reduce (scale down) the amplitude value of the operand.

That is, first, in the first step, a representative value is searched from index values in each unit.
As this representative value, for example, as described in claim 5, the maximum value in all index values is used.

Then, in the second step, whether or not to perform scaling is determined based on this representative value. Then, scaling is performed in the third step based on the determination result of this step.

Thus, according to the above method, it is possible to avoid an arithmetic error due to overflow by scaling and to decode the data having a minute amplitude with high accuracy. Furthermore, since it is determined whether or not to perform scaling, scaling is not performed on data that does not need to be scaled. Therefore, the amount of calculation in the decoding process can be reduced. Further, since this judgment is made based on the representative value of the index value in each unit, the processing amount for this judgment is very small. As described above, according to the above method, it is possible to improve the accuracy in decoding a digital signal without increasing the calculation amount.

Further, the digital signal according to claim 2 of the present invention, in addition to the method of claim 1, Spectra
IMDCT processing for performing IMDCT processing on system data
Further comprising the steps of:
Before processing, put it in blocks divided by frequency band.
Based on the maximum value of the scale factor of each unit
During the calculation process in the IMDCT process.
Scaling on operands to prevent bar flow
It is characterized by including the step of performing. Also, the present invention
The digital signal according to claim 3 is the signal according to claim 1 or
In addition to the method described in 2, it is characterized by including a fourth step of determining a scale amount for scaling in the third step based on the representative value.

According to the above method, all scale amounts of the arithmetic processing are determined based on the representative value. Therefore, the scale amount can be determined by a very simple process. Therefore, according to this method, the decoding method of the digital signal according to the first aspect can be easily realized, and the amount of calculation can be further reduced.

According to a fourth aspect of the present invention, in addition to the method according to the first or second aspect , the method for decoding a digital signal is characterized in that the scale amount for scaling in the third step is It is characterized in that it includes a fifth step of making a decision based on a value to be calculated in the calculation.

According to the above method, the scale amount in the arithmetic processing is determined based on the operated value which is the input data of each arithmetic processing. Therefore, in each calculation process, optimum scaling can be applied to the calculated value. Therefore, according to this method, the accuracy of decoding the digital signal can be made extremely high.

According to a fifth aspect of the present invention, in addition to the method according to the first or second aspect , the digital signal decoding method uses a scale amount for scaling in the third step as the representative value. It is characterized by including a sixth step of determining whether to determine based on the result of the determination, based on the result of the determination.

According to the above method, it is determined in the sixth step whether the scale amount is determined based on the representative value or the scale amount is determined in each calculation process based on the calculated value. It is like this. And
This judgment is made based on the representative value.

Therefore, according to this method, the method of determining the scale amount is determined according to the spectrum data to be decoded, so that the optimum scaling can be performed according to the characteristic of the spectrum data. There is. Furthermore, since this judgment is made using the representative value, the amount of calculation processing for this judgment is very small. Therefore, according to the above method, it is possible to perform optimal scaling without increasing the amount of calculation.

A decoding method for a digital signal according to a sixth aspect of the present invention is the method for decoding a digital signal according to any one of the first to fifth aspects, wherein the index value is a spectrum in each unit. In addition to the maximum amplitude value of the data, the representative value is a maximum value among the index values of all units.

According to the above configuration, the representative value becomes the maximum value in the amplitude value of the spectrum data. Therefore, since the scaling is performed based on this representative value, it is possible to reliably avoid the overflow in the arithmetic processing during the scaling. Further, in this method, since the scaling value and the maximum value of the index values only have to be selected in order to select the index value and the representative value, the selection process can be greatly simplified. As described above, according to the method described above, it is possible to reliably avoid the overflow in the calculation processing without increasing the calculation amount.

According to a seventh aspect of the present invention, there is provided a method for decoding a digital signal, wherein a unit is divided into a plurality of units according to a time range and a frequency domain, and each unit is divided based on an index value set for each unit. A method of decoding a digital signal for decoding spectrum data quantized in, comprising: a seventh step of searching a representative value from index values in each unit; The eighth step of scaling the value and the scale amount for scaling in the eighth step are determined based on the representative value or the calculated value in the calculation. Judge,
And a ninth step of determining the scale amount based on the result of the determination.

According to the above method, when the spectrum data divided into units is decoded, in order to avoid overflow in the arithmetic processing in the decoding process, or to decode the data of minute amplitude with high accuracy, The calculated value is scaled.

That is, first, in the seventh step, a representative value is searched from index values in each unit.
As the representative value, for example, as described in claim 7, the maximum value in all the index values is used. Then, in the ninth step, it is determined whether the scale amount is determined based on the representative value or the calculated value is determined based on the calculated value. Then, this judgment is made based on the representative value. Then, the scale amount is determined based on this determination result. Then, based on this scale amount, in the eighth step, the calculated value is scaled.

As a result, according to the above-mentioned method, it is possible to avoid arithmetic error due to overflow by scaling and to decode the data of minute amplitude with high accuracy. Furthermore, since the method of determining the scale amount is determined according to the spectrum data to be decoded, optimal scaling can be performed according to the characteristics of the spectrum data. Furthermore, since this judgment is made using the representative value, the amount of calculation processing for this judgment is very small. Thus, according to the above method, it is possible to decode a digital signal by optimal scaling without increasing the amount of calculation.

The decoding method for a digital signal according to claim 8 of the present invention is the method according to claim 7,
IMDCT processing for spectrum data
And the eighth step further comprises the IMDCT treatment.
Before, in blocks divided by frequency band,
Based on the maximum scale factor of each unit,
In the process of calculation processing in IMDCT processing
Scales the operands so that there are no rows
It is characterized by including steps. Also, the claims of the present invention
The method for decoding a digital signal according to claim 9 is the method according to claim 7.
Alternatively, in the method described in Item 8, the index value is the maximum amplitude value of the spectrum data in each unit, and the representative value is the maximum value in the index values of all the units.

According to the above method, the representative value becomes the maximum value in the amplitude value of the spectrum data. Therefore, since the scaling is performed based on this representative value, it is possible to reliably avoid the overflow in the arithmetic processing during the scaling. Further, in this method, since the scaling value and the maximum value of the index values only have to be selected in order to select the index value and the representative value, the selection process can be greatly simplified. As described above, according to the method described above, it is possible to reliably avoid the overflow in the calculation processing without increasing the calculation amount. Further, claim 1 of the present invention
The method of decoding a digital signal according to claim 0,
The method according to any one of 9 above,
The IMDCT treatment of the data and the IMDCT treatment.
Reverse to correct the amplitude value of the processed spectrum data
And a step of performing a floating process.
ing. Further, a digital device according to claim 11 of the present invention.
The method for decoding a signal is the method according to claim 10.
In the step of performing the reverse floating process,
Calculate the total sum of the scales used for the caring
The value obtained by multiplying the sum obtained by 3 times is used for each unit before IMDCT processing.
Off scale factor index in knit
Subtract from the set value to obtain a new offset value
Subtract the new offset value from the reference value
This is a new offset value.
The scale factor is the scale factor according to
Multiply the amplitude value of spectrum data that has been subjected to T processing
It is characterized by that. Moreover, in claim 12 of the present invention.
The described digital signal decoding method is based on the time range and frequency.
It is divided into several units by several areas and each unit
Encoding for each unit based on the index value set for each
Digit for decoding the spectrum data being
Tal signal decoding method
For the process of performing IMDCT processing, and for each block
Find the unit with the largest scale factor
The scale factor of this unit
Before the IMDCT process and the process of obtaining the table values,
In blocks divided into each frequency band, each unit
IMDC based on the maximum scale factor of
Overflow occurs in the process of arithmetic processing in T processing
So that there is no
The amplitude value of the spectrum data that has been subjected to MDCT processing
Including the step of performing a reverse floating process for correction,
In the process of performing the reverse floating process, the scale
The sum of the scale quantities used for
The value obtained by multiplying by 3 is used in the IMDCT processing step.
Previous scale factor index for each unit
New offset value
Calculate and subtract this new offset value from the reference value
To a new offset value,
The scale factor according to the offset value is
Then, of the spectrum data that has been IMDCT processed
It is characterized by multiplying the amplitude value.

[0034]

BEST MODE FOR CARRYING OUT THE INVENTION [First Embodiment] A first embodiment of the present invention will be described below. FIG. 2 is an explanatory diagram showing the configuration of the mini disk recording / reproducing apparatus 1 which is the digital recording / reproducing apparatus according to the present embodiment. As shown in this figure, the mini disk recording / reproducing apparatus 1 comprises an input terminal 2, a photoelectric element 3, a digital PLL circuit 4, a frequency conversion circuit 5, an audio compression circuit 6, an audio expansion circuit 9, a shock proof memory controller 7, a shock. Proof memory 8, signal processing circuit 10, head drive circuit 11, recording head 12, optical pickup 13, spindle motor 14,
Feed motor 15, driver circuit 16, servo circuit 17,
RF amplifier 18, D / A converter 19, output terminal 20, system control microcomputer 22, and power ON / OF
The F circuit 23 is provided. Further, the magneto-optical disk 25 in this figure is a mini disk capable of recording / reproducing in this mini disk recording / reproducing apparatus 1.

First, each component of the mini disk recording / reproducing apparatus 1 will be described. The photoelectric element 3 is for converting a digital signal of an optical signal input to the input terminal 2 into a digital signal of an electric signal. Digital PLL circuit 4
Extracts a clock from the digital signal output from the photoelectric element 3 and outputs 48 kH for this signal.
z, 44.1 kHz and 32 kHz are sampled, and a signal corresponding to the number of quantization bits is generated and output. Hereinafter, the digital signal generated by the digital PLL circuit 4 will be referred to as audio data.

The frequency conversion circuit 5 is for converting the sampling frequency of the audio data generated by the digital PLL circuit 4 into 44.1 kHz corresponding to the mini disk standard. The audio compression circuit 6 adds A to the audio data whose frequency is converted by the frequency conversion circuit 5.
The data is compressed and encoded by the TRAC (Adaptive TRansform Acoustic Coding) method and is output as audio data (encoded audio data). The configuration and operation of the audio compression circuit 6 will be described later.

The shock proof memory 8 is a memory for shock proof processing. The shock proof memory controller 7 utilizes the shock proof memory 8 to absorb the difference between the speed at which the audio compression circuit 6 outputs audio data and the processing speed of the signal processing circuit 10. Further, the shock proof memory controller 7 uses the shock proof memory 8 to output a sound input to the signal processing circuit 10 when the digital signal input to the input terminal 2 is interrupted by a disturbance such as vibration during reproduction. Data is interpolated to protect digital signals or audio data. Such processing by the shock proof memory controller 7 and the shock proof memory 8 is referred to as shock proof processing.

The signal processing circuit 10 has functions as an encoder and a decoder. That is, the signal processing circuit 10 inputs the audio data output from the shock proof memory 8 when recording a digital signal,
The audio data is subjected to encoding processing such as parity addition, and the audio data is modulated into a magnetic modulation signal that can be recorded on the magneto-optical disk 25 and output to the head drive circuit 11. When reproducing a digital signal, the signal processing circuit 10 performs a decoding process on the audio data reproduced from the magneto-optical disk 25 and outputs it to the audio expansion circuit 9 via the shock proof memory controller 7. .

The audio decompression circuit 9 is for decoding the audio data output from the signal processing circuit 10 by the ATRAC method. Then, the decoded voice data, that is, the audio data is transferred to the D / A converter 19
Is converted into an analog audio signal by and output from the output terminal 20. The configuration and operation of the voice decompression circuit 9 will be described later.

The head drive circuit 11 moves the recording head 12 to a predetermined recording position on the magneto-optical disk 25 and causes the recording head 12 to generate a magnetic field corresponding to the magnetic modulation signal output from the signal processing circuit 10. It is a thing.

The optical pickup 13 irradiates a predetermined recording position on the magneto-optical disk 25 with laser light during recording and reproduction. Then, at the time of reproduction, the reflected light of the laser light from the magneto-optical disk 25 is read to detect the RF signal (modulated audio data) recorded on the magneto-optical disk 25. The RF amplifier 18 amplifies the RF signal detected by the optical pickup 13. The spindle motor 14 is the magneto-optical disk 2
5 is rotated in a predetermined direction. Feed motor 1
5 is for moving the optical pickup 13 in a direction orthogonal to the track on the magneto-optical disk 25.

The driver circuit 16 includes a magneto-optical disk 25.
The electric power is supplied to the feed motor 15, the spindle motor 14, and a drive device (not shown) that drives an objective lens (not shown) of the optical pickup 13 so that the laser beam is irradiated to a predetermined recording position of the above. Servo circuit 1
Reference numeral 7 is based on the output of the RF amplifier 18 so that an operation for recording / reproducing such as causing the laser light emitted from the optical pickup 13 to follow a target track of the magneto-optical disk 25 is accurately performed. The feedback control is performed on each device driven by the driver circuit 16.

The power ON / OFF circuit 23 is for supplying electric power to the signal processing circuit 10, the optical pickup 13, the driver circuit 16, the servo circuit 17, the RF amplifier 18, the system control microcomputer 22, and the like.
The system control microcomputer 22 turns on / off the power.
The central part of the mini disk recording / reproducing apparatus 1 is for managing the power ON / OFF operation of the F circuit 23 and for centralized management of the signal processing operation by the above-mentioned members.

As described above, the voice compression circuit 6 and the voice decompression circuit 9 are coded by the ATRAC system.
It is a DSP in the mini-disc recording / reproducing apparatus 1 that performs decoding. The ATRAC method is a method used in a mini disk as a method of compressing and encoding a digital signal such as a musical sound or a voice with high efficiency. In this method, a digital signal sampled at 44.1 kHz is divided into a predetermined frequency band (low range of 0 to 5.5 kHz, middle range of 5.5 to 11 kHz, high range of 11 to 22 kHz). , Each frequency band is divided into blocks of variable length unit time. The blocked digital signal is converted into MDCT (modified discrete cosine).
Transform processing is performed to encode each predetermined frequency domain. This encoding is performed based on the number of bits assigned using the psychoacoustic characteristics.

Next, the voice compression circuit 6 will be described.
FIG. 3 is an explanatory diagram showing the configuration of the audio compression circuit 6. As shown in this figure, the voice compression circuit 6 includes a band division filter unit 31, a time window cutout unit 32, an MDCT unit 33, a bit allocation unit 34, a unit division unit 35, and a quantization unit 3.
6 and the encoding unit 37.

First, each component of the audio compression circuit 6 will be described. The band division filter unit 31 uses a QMF (quadratu).
re-mirror filter) etc., and the audio data input from the frequency conversion circuit 5 is divided into a high-frequency side and a low-frequency side.
It is divided into two frequency bands. The time window cutout unit 32 divides the input audio data into units of a predetermined unit time (11.6 milliseconds, 512 samples).

In the following, the audio data sectioned by the time window cutout unit 32 for each unit time will be referred to as grouped audio data. In addition, the audio data divided into unit times and divided by the band division filter unit 31 for each frequency band is set as blocked audio data. That is, the blocked audio data is divided into groups on the time axis,
It is audio data divided into blocks on the time axis and the frequency axis.

The MDCT unit 33 performs MDCT (modifi) on a block-by-block basis for the block-shaped audio data.
ed discrete cosine transform: modified discrete cosine transform). Since the MDCT unit 33 performs MDCT processing on the audio data for each block, the spectrum data generated after the processing is also divided into blocks.

The bit allocation unit 34 allocates the number of quantized bits, that is, the word length (WL), to the blocked spectrum data generated by the MDCT unit 33 using the psychoacoustic characteristics. The unit dividing unit 35 divides the blocked spectrum data generated by the MDCT unit 33 into units that are further divided by a plurality of frequency regions and time ranges. In the following, the spectrum data divided into units will be referred to as unitized spectrum data. Then, the maximum amplitude value in the spectrum data of each unit is obtained as the index value of each unit. Then, this index value is set as a scale factor (SF) of each unit.

The quantizing unit 36 quantizes the unitized spectrum data for each unit based on the word length and scale factor of each unit calculated by the bit allocating unit 34 and the unit dividing unit 35. .

The encoding unit 37 is capable of recording the spectrum data quantized by the quantization unit 36 on the magneto-optical disk 25 shown in FIG. 212 bytes; 1
Sound group) is compressed (encoded) to generate audio data in a predetermined compression format. The audio data is spectrum data that has been quantized and compressed (encoded). The compression format of this audio data will be described later.

Next, the audio data compression operation by the audio compression circuit 6 will be described. Frequency conversion circuit 5
Sampling frequency 4 output from (see Fig. 2)
When the 4.1 kHz audio data is input to the voice compression circuit 6, the band division filter unit 31 divides the audio data into two blocks, that is, a high frequency band and a low frequency band. And
The audio data of the low-frequency side block is input to the band division filter unit 31 again, and the frequency band is divided into two. As a result, the audio data input to the audio compression circuit 6 is divided by the band division filter unit 31 into three frequency bands of high band, middle band, and low band.

The audio data divided into three bands by the band division filter unit 31 is divided by the time window cutout unit 32 into a maximum time window of 11.6 ms (512 samples). Therefore, the audio data is 5
It can be said that they are grouped by a time window of 12 samples and are divided into three frequency bands within the group. In this way, the audio data becomes blocked audio data divided into a time window of 512 samples and three frequency bands.

Thereafter, the blocked audio data is MDCT processed by the MDCT unit 33 for each block.
It is processed and converted into blocked spectrum data. Then, after the bit allocation section 34 performs bit allocation and word length calculation on the blocked spectrum data, the unit division section 35 generates unitized spectrum data and The scale factor in the unit is set.

FIG. 4 is an explanatory diagram showing an example of unitized spectrum data according to one group of audio data, using a time axis and a frequency axis. As shown in this figure, the low-frequency block is divided into four equal parts on the time axis and five parts on the frequency axis, so that 20 units (Unit in FIG. 4) can be obtained.
It is divided into 0 to Unit 19). Each unit in the low-frequency block has eight pieces of spectrum data (for example, AS in Unit 0 in FIG. 4).

[0] to AS [7]) are included.

The mid-range block is divided into four equal parts on the time axis and on the frequency axis, so that 16 units (Unit in FIG. 4) can be obtained.
20-Unit35). Then, each unit in the mid-range block has six pieces of spectrum data (for example, AS [1 in Unit 20 in FIG. 4).
28] to AS [133]) are included.

The high-frequency block is divided into eight equal parts on the time axis and two equal parts on the frequency axis, so that 16 units (Unit in FIG. 4) can be obtained.
36 to Unit 51). 12 units of spectrum data (for example, AS in Unit 36 in FIG. 4) are included in each unit in the high frequency block.
[256] to AS [267]) are included. Thus, in this example, 512 (AS

[0] to AS [51
1]) has 52 pieces of spectrum data (Unit0 to Unit0).
It is divided into units of Unit 51). Moreover, the number of spectrum data included in each unit is different for each of the three blocks. The method of taking the blocks and units shown in FIG. 4 is an example, and the present invention is not limited to this.

After being unitized by the unit division unit 35 as described above, the spectrum data is quantized by the quantization unit 36 based on the word length and the scale factor. Then, the quantized spectrum data is 424 bytes (2 × 212 bytes; 2 × 212 bytes;
It is converted into audio data of one sound group).

FIG. 5 is an explanatory diagram showing the format of this audio data. Note that FIG. 5 shows one channel (212 bytes) of the 424 bytes.
As shown in this figure, in this format, main data BSM, SIA, word length index and scale factor index, quantized spectrum data, sub data BSM, SIA, word length index and scale The index of the factor is recorded. The content of the information of the sub data is almost the same as the information of the main data.

The BSM (block size mode) is information on blocking and unitizing the spectrum data. In addition, SIA (sub information amount)
Is information on the number of units in each block of spectrum data. Word length index (WLI)
Is the word length (W
L) is the index. Table 1 shows the correspondence between these WLI and word length (WL). As shown in this table, the word length (WL) is an integer from 0 to 16 excluding 1, and WLIs from 0 to 15 correspond in ascending order.

[0061]

[Table 1]

Scale factor index (SF
I) is an index of the scale factor (SF), which is the magnitude of the amplitude of the representative spectrum data of each unit. Table 2 shows the correspondence between these SFIs and scale factors (SF). As shown in this table, the scale factor (SF) is a number represented in the form of a product of the coefficient m [i] and 2 ⁿ . Where i and n are 1
It is an integer satisfying ≦ i ≦ 3, −5 ≦ n ≦ 16. Also, m
[I] is m [1] = 0.62996052, m [2] = 0.793700
52, m [3] = 0.99999999. Also, n = -5 and m
The product of [1] and m [2] is excluded from the scale factor.

The SFIs of 0 to 63 correspond to the scale factor in ascending order. That is,
The scale factor has 64 steps. Also, three coefficients m [1], m [2], m [3] for one n.
1 bit corresponds to three SFIs.

[0064]

[Table 2]

The scale factor in the block is
There is a maximum of 5 because it exists for each unit included in each block.
There are two. On the other hand, the maximum number of quantized spectrum data is 512 because there are the same number of sampled audio data. The unit also contains at least one piece of quantized spectrum data. Therefore, it is clear that the number of scale factors in a block is smaller than the number of quantized spectrum data.

Next, the audio decompression circuit 9 shown in FIG. 2 will be described. FIG. 6 is an explanatory diagram showing the configuration of the audio decompression circuit 9. As shown in this figure, the voice expansion circuit 9
Is a storage unit 40, a WL expansion unit 41, an SF expansion unit 42, A
The S expansion unit 43, the SF maximum value determination unit 44, the SF offset attaching unit 45, the inverse quantization unit 46, the IMDCT unit 47, the inverse floating unit 48, the windowing unit 49, and the band synthesis filter unit 50 are provided.

The WL expansion unit 41 expands the word length in each unit in each of the three blocks into the actual word length value based on the WLI in the audio data of the format shown in FIG. To do. Further, the SF expansion unit 42 similarly expands the scale factor in each unit in the block according to Table 2 based on the SFI.

The AS decompression unit 43 decompresses the spectrum data compressed over several words, word by word. That is, the AS expansion unit 43 reads out the quantized spectrum data stored in various states by 8 bits and expands it for each word.

Since the maximum value of the word length is 16 bits, if one word can be read by one reading,
In some cases, reading can be performed by reading twice, and reading can be performed by reading three times. That is, the processing content of the AS expansion unit 43 is to read out the quantized spectrum data stored in various states by 8 bits and cut it by 1 word (maximum 16 bits).

The SF maximum value determination unit 44 searches the unit having the maximum scale factor in each block and acquires the scale factor of this unit as the representative value of the block. The SF offset attaching unit 45 updates the SFI of each unit for each block so that the maximum value of the scale factor acquired by the SF maximum value determining unit 44 becomes m [3] × 2 ¹⁶ (SFI = 63). To do. That is, the SF offset attaching unit 45
Is the SF according to the maximum scale factor of each block
I is subtracted from 63 and set as an SFI offset value (scale amount). Then, the SF offset attaching section 45 updates the SFIs by adding the obtained offset values to the SFIs of all the units in the block. This offset value is SFI
Is updated so that the processing of the IMDCT unit 47 described later does not overflow.

The dequantization unit 46 dequantizes the spectrum data in the voice data based on the word length obtained by the WL expansion unit 41 and the scale factor indicated by the SFI updated by the SF offset attaching unit 45. Is to do. This dequantization will be described later.

The IMDCT section 47 performs an IMDCT on the spectrum data dequantized by the dequantization section 46.
(Invers e modified discrete cosine transform) processing performed, and generates the audio data blocks. The detailed configuration and processing of the IMDCT unit 47 will be described later.

The inverse floating section 48 appropriately corrects and outputs the amplitude value of the blocked audio data generated by the IMDCT section 47. That is, in the IMDCT processing in the IMDCT unit 47, scaling is performed on the spectrum data in order to prevent overflow of the arithmetic processing. Therefore, the inverse floating unit 48 is adapted to correct the amplitude value converted by this scaling to an appropriate value. The scaling means the IMDCT unit 4
In the IMDCT process in 7, the scale down and scale up are performed on the data (calculated value) related to the calculation.

The window portion 49 is the reverse floating portion 48.
Blocked audio data output from
They are connected on the time axis. That is, the window portion 49
Is a windowing process between certain data and data one sample before on the time axis.

The band synthesizing filter unit 50 uses the IQMF (In
vers quadrature mirror filter) and other band synthesis filters, and the audio data divided into three for each frequency band are combined into one. That is, the band synthesizing filter unit 50 synthesizes the blocked audio data and outputs it to the D / A converter 19 shown in FIG. The storage unit 40 includes the voice expansion circuit 9 described above.
Each of the configurations is for storing information necessary for each process.

Next, the operation of expanding the audio data by the audio expansion circuit 9 will be described. Note that (2) to be described later
IMDCT processing using equation (6) is described in IEICE Technical Report, CAS90-
9 DSP90-13, “A Study on MDCT Method and High-speed Arithmetic”, Masahiro Iwadare, Takao Nishitani, Akihiko Sugiyama, et al., And I
EEE, Transactions on ASSP, Vol. 34, No. 5, "Analysis / Synthesis Filterbank Design Based on Aliasing Cancellation in Time Domain", John P. Princen, Alan Bernard Br
See adley for details.

When the audio data of the format shown in FIG. 5 is input to the audio decompression circuit 9 from the signal processing circuit 10 via the shock proof memory controller 7 shown in FIG. Based on this, the word length of each unit in the three blocks of the group corresponding to the audio data is expanded and stored in the storage unit 40. Similarly, the SF expansion unit 42 expands the scale factor of each unit based on the SFI and stores it in the storage unit 40.

After that, the AS expansion unit 43 reads out the quantized spectrum data in the audio data by 8 bits and expands it by each word. Then, the SF maximum value determination unit 44 searches the unit having the maximum scale factor for each block, and acquires the maximum value of the scale factor in each block. Then, the SF offset attaching unit 45 calculates the SFI offset value (scale amount) so that the maximum value of the scale factor in each block is m [3] × 2 ¹⁶ (SFI = 63).

Then, the SF offset attaching section 45 updates the SFI of each unit by adding this offset value to the SFI of all the units belonging to each block. The SF offset attaching unit 45 also calculates the calculated offset value, the updated SFI, and this SF.
The scale factor corresponding to I is stored in the storage unit 40.

After the SFI of each block is updated, the inverse quantization unit 46 calculates the word length expanded by the WL expansion unit 41,
The scale factor according to the SFI updated by the SF offset assigning unit 45 is acquired from the storage unit 40. Then, using these values and the following equation (1), inverse quantization processing is performed for each unit on the quantized spectrum data expanded by the AS expansion unit 43. This allows
The quantized spectrum data AS [j] is inversely quantized to generate unitized spectrum data X [k].

[0081]

[Equation 1]

In this equation, WL [i] indicates the word length of each unit, i indicates the unit number, j indicates the quantized spectrum data number, and k indicates the spectrum data number. In addition, 0 ≦ i ≦ N, 0 ≦
j, k ≦ M−1. N indicates the maximum value of the unit number, and M-1 indicates the maximum value of the spectrum data number. Also, N <M-1.

After that, the IMDCT unit 47 makes an I for the unitized spectrum data for each block.
MDCT processing is performed to generate audio data blocked in three frequency bands of high frequency band, middle frequency band, and low frequency band, and output to the inverse floating unit 48. In addition, IMD
The detailed configuration and operation of the CT unit 47 will be described later.

After that, the inverse floating unit 48 uses the predetermined SFI (for example, 63) as a reference, and the IMDCT unit 4
The amplitude value of the blocked audio data output by 7 is corrected and output to the windowing unit 49, and the reverse floating is completed. The operation of the reverse floating unit 48 will be described later.

After that, the windowing section 49 performs windowing processing on the blocked audio data output from the inverse floating section 48 and adjusts between adjacent frames on the time axis to obtain a band synthesis filter. Output to the unit 50. Then, the band synthesis filter unit 50 generates one group of audio data, and the decoding processing by the audio decompression circuit 9 ends.

Next, the configuration and operation of IMDCT section 47 will be described. FIG. 7 is an explanatory diagram showing the configuration of the IMDCT unit 47. As shown in this figure, the IMDCT unit 47 includes a U (k) calculation unit 51, an FFT calculation unit 52, u.
An (n) calculation unit 53, a y (n) calculation unit 54, a fixed scaling unit 55, a fixed scale amount loading unit 56, and a fixed scale amount memory 57 are provided.

As described above, the IMDCT unit 47 performs IMDCT processing on the unitized spectrum data.
T processing is performed. That is, the U (k) calculation unit 5
The 1-y (n) calculator 54 performs the arithmetic processing shown in the following (2)-(6) on the unitized spectrum data.

Then, the fixed scaling unit 55, the fixed scale amount loading unit 56, and the fixed scale amount memory 57.
Is to prevent overflow in the process of calculation before the calculation process shown in the equations (3) to (5) is performed.
A predetermined scaling is performed on the calculated value.

The operation of IMDCT section 47 will be described below. When the unitized spectrum data is input to the IMDCT unit 47, the U (k) calculation unit 51
U (k) corresponding to the spectrum data X [k] input to the IMDCT unit 47 is calculated using the following equation (2), and is output to the fixed scaling unit 55. The fixed scaling unit 55 performs predetermined scaling described later on this U (k) and outputs it to the U (k) calculation unit 51.

After that, the U (k) calculator 51 derives Z (L) in each U (k) using the following equation (3) and outputs it again to the fixed scaling unit 55. Note that Re [Z (L)] and Im [Z (L)] in this equation are
The real and imaginary parts of the complex number Z [L] are shown. Also, L is
The number is 0 ≦ L <M / 2.

[0091]

[Equation 2]

[0092]

[Equation 3]

The fixed scaling unit 55 receives the input Z
The predetermined scaling described later is performed on (L), and F
Input to the FT calculation unit 52. The FFT calculation unit 52 performs a fast Fourier transform on each input Z (L) using the following equation (4). That is, the FFT calculation unit 52
Is provided with an FFT first-stage calculation unit to an FFTP-stage calculation unit with P = log ₂ (M / 2), and performs fast Fourier transform by a P-stage butterfly operation to calculate z (n).
Note that i in the equation (4) is an imaginary unit. Also, n
Is a number satisfying 0 ≦ n <M / 2.

In this calculation, each FFTp stage calculation unit (p
= 1 to P), the calculated result is output to the fixed scaling unit 55. Then, the fixed scaling unit 55 performs a predetermined scaling on the input calculation result and outputs the result to the FFT (p +
1) It is designed to be output to the stage calculation unit. And
After the P stage butterfly calculation is completed, the FFT calculation unit 52
The z (n) is output to the fixed scaling unit 55.

[0095]

[Equation 4]

The fixed scaling unit 55 receives the input z
(N) is subjected to predetermined scaling described later,
It is output to the u (n) calculator 53. u (n) calculator 53
Calculates u (n) using the following equation (5) based on the input z (n), and outputs it to the y (n) calculator 54. After that, the y (n) calculator 54 receives the input u
Based on (n), using the following equation (6), the IMDCT
Audio data y that has been processed and blocked
Calculate (n).

[0097]

[Equation 5]

[0098]

[Equation 6]

Here, the fixed scaling unit 55 described above is used.
The scaling in will be described. The fixed scale amount memory 57 of the IMDCT unit 47 stores U (k), F
A fixed scale amount is stored for an input value (including Z (L)) to the FTp-th stage calculation unit and a calculated value that is preferably scaled, such as z (n).

Then, the fixed scaling unit 55 controls the fixed scale amount loading unit 56, and based on the maximum scale factor of each block, obtains the fixed scale amount according to the type of each operation to be performed. It is read from the fixed scale amount memory 57. Then, based on the read fixed scale amount, the calculated values are sequentially scaled.

Table 3 shows the fixed scale amount stored in the fixed scale amount memory 57. As shown in this table, these fixed scale quantities are
Scale factor (S
It corresponds to the maximum value of F).

In this table, the fixed scale amount is represented by the number of bits. That is, the positive integers in this table are
It is a scale amount for performing scaling for bit-shifting the operand value to the LSB (Least Significant Bit) side (hereinafter referred to as the right side). Further, the negative integer is a scale amount for performing scaling for bit-shifting the operand value to the MSB (Most Significant Bit) side (hereinafter, left side). In this table, the input values to the FFT calculation unit are shown up to the fourth level. Also, in this table,
The values at the maximum scale factor values of 2 ⁸ to 2 ⁻⁴ are omitted, because the fixed scale amounts at the maximum scale factor values of 2 ⁹ or less are all the same value.

[0103]

[Table 3]

The fixed scale amount shown in this table is
It is calculated as follows. That is, the above (3)-
The calculation result obtained by the calculation processing shown in the equation (5) is different in magnitude from the calculated value. Table 4 shows the multiplication factor of the calculation result with respect to the calculated value for each calculation process. A margin bit as shown in Table 4 is generated in the upper bits of the calculated value of the calculation result, depending on the ratio (or magnification) of the calculated value before and after the calculation. This margin bit is used by the IMDCT unit 4
7 design. Table 5 shows the ratio of the calculated value to the calculation result for each calculation process. The method for obtaining the magnifications and ratios shown in Tables 4 and 5 is described in detail in JP-A-7-36666.

[0105]

[Table 4]

[0106]

[Table 5]

Therefore, the calculation result becomes too large and I
It can be seen that it is preferable to set the fixed scale amount so that the processing in any configuration in the MDCT unit 47 does not overflow. Then, the fixed scale amount is determined based on the following conditions.

The maximum amplitude value of the spectrum data input to the IMDCT unit 47 is 2 ¹⁶ . The maximum amplitude value of the audio data output from the IMDCT unit 47 is 2 ¹⁵ . Depending on the above and the magnifications / ratios shown in Tables 4 and 5, the calculation result in each configuration of the IMDCT unit 47 has a maximum value.

When the maximum value of the calculation result in each component of the IMDCT unit 47 is represented by a power of 2, the fixed scale amount shown in Table 3 can be obtained by obtaining the envelope curve representing each maximum value. it can. The method for obtaining the fixed scale amount is described in detail in the above-mentioned Japanese Patent Laid-Open No. 7-36666.

Next, the operation of the reverse floating section 48 will be described. As described above, the reverse floating portion 4
Reference numeral 8 corrects the amplitude value of the audio data output from the IMDCT unit 47. That is, the inverse floating unit 48 obtains the total sum of the fixed scale amounts used for the scaling performed on the operands before the processing of the above equations (3) to (5), and obtains the obtained total sum. The tripled value is subtracted from the offset value of the SFI in each unit before the processing in the IMDCT unit 47 to newly obtain the offset value.

Then, based on this offset value, the amplitude value of the audio data is corrected using a predetermined SFI as a reference value. That is, the inverse floating unit 48 subtracts the obtained offset value from the reference value to obtain a new offset value. Then, the scale factor corresponding to the offset value is used as the scale amount, and the amplitude value of the audio data input from the IMDCT unit 47 is multiplied. As a result, the reverse floating process by the reverse floating unit 48 ends. The inverse floating process is a process in which the IMDCT process is performed and the audio data output from the IMDCT unit 47 is multiplied by a scale factor according to a predetermined offset value.

Here, the reverse floating portion 48 is
The reason why a value obtained by multiplying the total sum of fixed scale amounts by 3 is used when I is updated will be described. The scale factor of the spectrum data is set at intervals of 1 dB divided into three equal parts. That is, the unit of 1 bit corresponds to 3 ranges in SFI (1 bit = 3 indexes). Therefore,
The inverse floating unit 48 subtracts a value obtained by multiplying the sum of the numbers of bits of the fixed scale amount by 3 from the offset value in order to return the SFI offset value.

An example of the processing of the audio decompression circuit 9 after the acquisition of the maximum scale factor value by the SF maximum value determination unit 44 will be described below. The maximum value of the scale factor of all units in a block of audio data for decoding is m [1] × 2 ¹⁰ = 0.62996052 × 2 ¹⁰ (SFI = 4
3). The SF offset attaching unit 45 determines that this value is SFI = 63 (m [3] × 2 ¹⁶ = 0.99999999 × 2
¹⁶ ) Calculate the offset value (scale amount) so that That is, the offset value in this case is 63-4
3 = 20.

Then, for example, in this block, the scale factor is m [2] × 2 ⁶ = 0.79370052 × 2 ⁶ (S
Suppose there is a unit with FI = 32). The SF offset attaching section 45 updates the SFI by adding the above scale amount as an offset value to the SFI of this unit. That is, the updated SFI is 32 + 20 = 52
(SF = m [1] × 2 ¹³ = 0.62996052 × 2 ¹³ )
In this way, the SF offset attaching unit 45 updates the SFI in all units in the block.

After that, the inverse quantization unit 46 performs inverse quantization to obtain the spectrum data X from the above equation (1).
After [k] is generated, the IMDCT unit 47 determines that the maximum value of the scale factor in this block is SF = m [1] ×
Since it is 2 ¹⁰ , the maximum scale factor value in Table 3 is referred to the column of 2 ¹⁰ to read the fixed scale amount in each arithmetic processing. Then, before the arithmetic processing, after scaling the operand, each arithmetic processing of the IMDCT processing is performed.

That is, before the arithmetic processing in the U (k) calculation unit 51 shown in the equation (3), the calculated value is set to the right by 1
Only shift by bits. Then, F shown in equation (4)
Prior to the processing in the FT first stage calculation unit 52 to the FFT third stage calculation unit, the operands are each shifted to the right by one bit. Then, before the processing in the FFT fourth stage calculation unit, the operand value is shifted to the right by 2 bits. Then, before the processing in the u (n) calculation unit 53 shown in the equation (5), the bit shift is not performed.

After that, when the blocked audio data is generated and input to the inverse floating unit 48, the inverse floating unit 48 firstly transfers the IMDCT unit 4 to the IMDCT unit 4.
The total of the fixed scale amount in 7 is calculated. That is,
The maximum value of the scale factor in this block is m [1]
Since it was × 2 ¹⁰ , the maximum scale factor value in Table 3 was 2 ^{10 and} the fixed scale amount in the IMDCT unit 47 was read to obtain the total sum.
Multiply to get the SFI offset value.

From this table, the sum of fixed scale amounts is 1
+ 1 + 1 + 1 + 2 = 6, and the SFI offset value is 18. Further, since the offset value given from the SF offset attaching unit 45 is 20, the offset value is updated to 20− (6 × 3) = 2. Further, the inverse floating unit 48 uses SFI = 15 as a reference value,
Update the offset value. That is, 15-2 = 13. Thereafter, inverse floating unit 48 multiplies a scale factor m [1] × 2 ⁰ indicated by this offset value and the amplitude value of the audio data blocks.
Note that the multiplication may be performed with a fixed point.

As described above, the IMDCT section 4 in the audio decompression circuit 9 provided in the mini disk recording / reproducing apparatus 1.
In 7, in the blocks divided for each frequency band,
Based on the maximum scale factor of each unit, I
The operand value is scaled so that overflow does not occur in the course of each operation process in the MDCT process. As a result, it is possible to reliably remove the calculation error caused by the overflow.

Further, the IMDCT unit 47 is designed to search for a scale factor having a smaller number than the spectrum data in order to obtain the fixed scale amount. Therefore, it is possible to significantly reduce the amount of calculation for obtaining the fixed scale amount. Furthermore, since the scale factor is an absolute value, this calculation amount can be further reduced. This makes it possible to reduce power consumption for arithmetic processing.

In the mini disk recording / reproducing apparatus 1,
The spectrum data scale factor is 1 dB to 3
It is set at equal intervals. Therefore, the scaling by the IMDCT unit 47 can be performed in units more detailed than 1 bit. Therefore, it is possible to improve the accuracy of the arithmetic processing as compared with the configuration in which the scaling is performed in bit units.

In this embodiment, the IMDCT unit 4
The fixed scale amount stored in the fixed scale amount memory 57 in No. 7 is the value shown in Table 3. However, the fixed scale amount stored in the fixed scale amount memory 57 is not limited to that in this table. The fixed scale amount may be set so that overflow does not occur in the arithmetic processing of the U (k) calculation unit 51 to y (n) calculation unit 54 in the IMDCT unit 47.

The scaling by the IMDCT unit 47 and the inverse quantization by the inverse quantization unit 46 may be performed simultaneously. Since the minidisk recording / reproducing apparatus 1 performs the above processing by using the scale factor, it is possible to adopt such a configuration. In this way, the amount of calculation can be further reduced, so that the power consumption can be further reduced.

In this embodiment, the scale factor (index value) of each unit is the maximum amplitude value of the spectrum data in each unit, and the representative value of each block is the maximum value of the scale factor. . However, the method of selecting the index value and the representative value is not limited to this, and may be selected by a method desired by the user. Further, as the SFI reference value used by the inverse floating unit 48, a value desired by the user can be used.

Further, in the present embodiment, in the mini disk recording / reproducing apparatus 1, the audio compression circuit 6 and the audio decompression circuit 9 are adapted to perform ATRAC system signal processing. However, the invention is not limited to this, and instead of the audio compression circuit 6 and the audio expansion circuit 9, an audio compression / expansion circuit provided with an ATRAC processing circuit may be provided. Then, the ATRAC processing circuit in this audio compression / expansion circuit may perform all the processes of the audio compression circuit 6 and the audio expansion circuit 9.

Further, in the voice expansion circuit 9 of this embodiment,
Although predetermined scaling is performed in each processing, it is also possible to adopt a configuration in which scaling is not performed. FIG. 10 is an explanatory diagram showing the configuration of the voice decompression circuit that is not scaled. This configuration has the SF maximum value determining unit 44 and the SF offset attaching unit 4 shown in FIG.
5, instead of the IMDCT unit 47 and the reverse floating unit 48, an IMDCT unit 81 is provided.

FIG. 11 is an explanatory diagram showing the structure of the IMDCT unit 81. As shown in this figure, the IMDCT unit 81 is a configuration that does not include the fixed scaling unit 55, the fixed scale amount loading unit 56, and the fixed scale amount memory 57 in the configuration of the IMDCT unit 47 shown in FIG. 7.

The IMDCT unit 81 is the same as the IMDCT unit 47 except that scaling is not performed. Further, the voice decompression circuit including the IMDCT unit 81 is the same as the voice decompression circuit 9 except that scaling is not performed. That is, when it is not necessary to perform scaling, it is possible to use the audio decompression circuit having such a configuration.

[Second Embodiment] A second embodiment of the present invention will be described below. In addition, the first embodiment described above
The members having the same functions as the members shown in (4) are designated by the same reference numerals, and the description thereof will be omitted.

The digital recording / reproducing apparatus according to the present embodiment is the same as that of the audio compression circuit 9 shown in FIG.
This is a configuration including an IMDCT unit 61 instead of the MDCT unit 47.

FIG. 8 is an explanatory diagram showing the structure of the IMDCT unit 61. As shown in this figure, the IMDCT unit 6
In the configuration of the IMDCT unit 47 shown in the first embodiment, 1 is a variable scaling unit 62, a variable scale amount determining unit 63, and a fixed scale amount loading unit 56 and a fixed scale amount memory 57, instead of the fixed scaling unit 62, the variable scale amount determining unit 63, and the variable scale amount determining unit 63. This is a configuration including a variable scale amount memory 64.

The IMDCT unit 61, like the IMDCT unit 47, performs IMDCT processing for each block on the spectrum data input from the inverse quantization unit 46. That is, U in the IMDCT unit 61
The (k) calculation unit 51 to y (n) calculation unit 54 performs the arithmetic processing shown in the above (2) to (6) on the input spectrum data.

Then, the variable scaling unit 62, the variable scale amount determining unit 63 and the variable scale amount memory 64.
In order to prevent overflow in the process of calculation, a predetermined scaling is performed on the calculated value before the calculation process shown in the equations (3) to (5) is performed.

The scaling of the calculated value by the variable scaling unit 62 will be described below. The variable scaling unit 62, like the fixed scaling unit 55, temporarily inputs the calculated value before the calculation processing shown in the equations (3) to (5) is performed. Then, the variable scaling unit 62 controls the variable scale amount determination unit 63 to determine the scale amount (variable scale amount) according to the input calculated value as follows.

That is, the variable scale amount determination unit 63
First, the maximum absolute value of the calculated value input to the variable scaling unit 62 is obtained. And the variable scale amount memory 6
Then, the magnifications stored in No. 4 in the arithmetic processing shown in Table 4 are read out. Then, the variable scale amount determination unit 63
Determines a variable scale amount for the calculated value so that the calculation process does not overflow even if the read magnification is multiplied by the maximum absolute value of the calculated value, and transmits the variable scale amount to the variable scaling unit 62.

If the maximum absolute value of the calculated operands is small, the variable scale amount determining unit 63 sets the variable scale amount to a value that scales up the operands to the extent that the arithmetic processing does not overflow. To decide. Then, the variable scaling unit 62 scales the calculated value based on the transmitted variable scale amount.

As described above, the IMDCT unit 61 determines the variable scale amount in the calculated value for each calculation process based on the magnification in Table 4 and the maximum absolute value of the calculated value to be input. Has become. Then, when the maximum absolute value is small, that is, when the operand is data of small amplitude, the operand is scaled up.

Therefore, it is possible to improve the accuracy of each calculation process when the calculated value is data of small amplitude. Further, since the variable scale amount is determined by focusing only on the maximum absolute value, not all the data of the calculated values, the calculation amount for determining the variable scale amount is shown in the first embodiment. The calculation amount of the IMDCT unit 47 is almost the same. Therefore, the IMDCT unit 61
Is capable of determining the variable scale amount with almost the same power consumption as the IMDCT unit 47.

[Embodiment 3] A third embodiment of the present invention will be described below. In addition, the first embodiment described above
A member having the same function as the member shown in and
The same reference numerals are given and the description thereof is omitted. The digital recording / reproducing apparatus according to the present embodiment has a configuration including an IMDCT section 71 in place of the IMDCT section 47 of the audio compression circuit 9 in the configuration of the mini disk recording / reproducing apparatus 1 shown in the first embodiment. .

FIG. 9 is an explanatory diagram showing the structure of the IMDCT unit 71. As shown in this figure, the IMDCT unit 7
1 includes a scaling unit 72 in place of the fixed scaling unit 55 in the configuration of the IMDCT unit 47 shown in the first embodiment, and the variable scale amount determining unit 63 and the variable scale amount memory 64 shown in the second embodiment. It is a configuration provided with.

The IMDCT unit 71, like the IMDCT unit 47, performs IMDCT processing for each block on the spectrum data input from the inverse quantization unit 46. That is, U in the IMDCT unit 71
The (k) calculation unit 51 to y (n) calculation unit 54 performs arithmetic processing shown in the following (2) to (6) on the input spectrum data.

Then, the scaling unit 72, the fixed scale amount loading unit 56, the fixed scale amount memory 57, the variable scale amount determining unit 63, and the variable scale amount memory 64.
In order to prevent overflow in the process of calculation, a predetermined scaling is performed on the calculated value before the calculation process shown in the equations (3) to (5) is performed.

The scaling of the calculated value by the scaling unit 72 will be described below. FIG. 1 is a flowchart showing a flow of operations in the scaling unit 72. As shown in this figure, the scaling unit 72
Inputs the spectrum data input to the U (k) calculator 51 (S1). Then, the scaling unit 72
Calculates the maximum absolute value of the scale factor in the input spectrum data (S2).

Then, the scaling unit 72 determines whether the scale amount to be used is fixed or variable based on the obtained maximum absolute value (S3.S).
5). That is, when the calculated maximum absolute value is equal to or greater than the preset first threshold value (for example, 2 ¹⁵ ), the fixed scale amount loading unit 56 is controlled in the same manner as the fixed scaling unit 55, and Table 3 Read the fixed scale amount shown in. That is, this fixed scale amount is used for the calculation in the IMDCT unit 71 (S4).

If the obtained maximum absolute value is smaller than the above-mentioned first threshold value and is equal to or more than the preset second threshold value (for example, 2 ³ ) or more, the scaling section 72 determines the variable scaling section 62. Similarly, the variable scale amount determining unit 63 is controlled to determine the variable scale amount for the calculated value for each calculation process. IMD
A variable scale amount is used for the calculation in the CT unit 71 (S6).

Further, the scaling section 72 does not perform scaling on the operand when the calculated maximum absolute value is smaller than the above-mentioned second threshold value. That is, I
No scaling is performed in the calculation in the MDCT unit 71 (S7). Then, after S4, S6 or S7, the blocked audio data is output (S
8) The processing by the IMDCT unit 71 ends.

As described above, the scaling unit 72 in the IMDCT unit 71 is configured to switch the processing for determining the scale amount according to the value of the scale factor of the spectrum data. As a result, the IMDCT unit 71
It is possible to further improve the accuracy of the IMDCT processing in (1) in comparison with the accuracy of the conversion processing in the IMDCT section 47 and the IMDCT section 61.

Further, the scaling in the scaling unit 72 is the process in the fixed scaling unit 55 or the variable scaling unit 62 added with only the process of determining the maximum value of the scale factor of the spectrum data using a threshold value. . Therefore, the difference between the calculation amount in the scaling unit 72 and the calculation amount in the fixed scaling unit 55 and the variable scaling unit 62 is negligibly small. Therefore, the amount of calculation and power consumption of the IMDCT unit 71 are the same as those of the IMDCT unit 47 and IM.
The amount is almost equal to that of the DCT unit 61. The first and second thresholds are determined according to the calculation amount and calculation accuracy desired by the user.

A program for performing all or a part of the processing in the IMDCT unit 71 shown in FIG. 1 has a CD-ROM (Read Only Memory) or an FD (Flop).
It is also possible to use an information processing device which is recorded in a recording medium such as a py disk) and which can read this program and which can input a digital signal from the outside, instead of the IMDCT unit 71.

Further, the scaling unit 72 described above is
(K) The maximum value of the spectrum data input to the calculation unit 51 may be obtained, and it may be determined whether the scale amount is fixed or variable based on the maximum value.

Further, the IMDCT section 47 and the IMDCT section 6
1 and the scaling performed in the processing of the IMDCT unit 71, the scale amount may be set according to the dynamic range accuracy of the scale factor.

The margin bits shown in Table 4 are generated in the higher order bits of the data after the operation depending on the data ratio of the operation data in the operation processing shown in the equations (3) to (5). It may be.

As described above, according to the first digital signal decoding method of the present invention, the digital input signal is orthogonally transformed into frequency domain spectrum data at every unit time, and the spectrum data is converted into a frequency domain unit. The spectrum data of each unit is quantized and encoded based on the representative value of the spectrum data of the unit to which it is divided, and then the quantized spectrum data and the representative value of the spectrum data of each unit are at least In a digital data decoding method for decoding encoded data recorded in a provided format, a representative value of spectrum data of each unit is searched, and based on the representative value, inverse quantization and inverse orthogonal transform are performed. To prevent overflow in the process, the scale is fixed by a predetermined fixed scale amount. How to pull up and scale down, and how to scale up and scale down by a variable scale amount corresponding to the operand data,
This is a method of selecting and using.

A second method for decoding a digital signal according to the present invention is the same as the first method for decoding a digital signal.
This is a method of searching the maximum value of the spectrum data representative value of each unit and performing it based on the maximum value.

According to the first and second digital signal decoding methods, scale-up and scale-down are performed by a predetermined fixed scale amount based on the representative value of the spectrum data of each unit, or By selecting whether to perform scale-up and scale-down according to the variable scale amount according to the operation data, the amount of decoding operation can be reduced compared to the conventional method of performing scale-up and scale-down using only the variable amount. Can be reduced. Further, it is possible to improve the calculation accuracy of decoding as compared with the conventional method of performing scale-up and scale-down using only a fixed scale amount.

The third digital signal decoding method of the present invention is the same as the first digital signal decoding method, in which a representative value of the spectrum data of each unit is searched and based on the representative value. , A method of scaling up and down by a predetermined fixed scale amount and a method of not scaling up and down so that overflow does not occur in the process of inverse quantization and inverse orthogonal transform, This is the method used.

A fourth method for decoding a digital signal according to the present invention is the same as the third method for decoding a digital signal.
This is a method of searching the maximum value of the spectrum data representative value of each unit and performing it based on the maximum value.

According to the third and fourth digital signal decoding methods, scale-up and scale-down are performed by a predetermined fixed scale amount based on the representative value of the spectrum data of each unit, or By selecting whether to perform scale-up and scale-down, it is possible to reduce the amount of decoding calculation as compared with the conventional method of performing scale-up and scale-down using only a fixed scale amount.

The fifth digital signal decoding method of the present invention is the same as the first digital signal decoding method, in which the representative value of the spectrum data of each unit is searched and based on the representative value. , A method of scaling up and down with a variable scale amount according to the data to be processed and a method of not performing scale up and down so that overflow does not occur in the process of inverse quantization and inverse orthogonal transformation This is the method used.

A sixth method for decoding a digital signal according to the present invention is the same as the fifth method for decoding a digital signal.
This is a method of searching the maximum value of the spectrum data representative value of each unit and performing it based on the maximum value.

According to the fifth and sixth digital signal decoding methods, whether the scale-up or scale-down is performed by the variable scale amount according to the data to be calculated based on the representative value of the spectrum data of each unit. Alternatively, by selecting whether or not to perform scale-up and scale-down, the amount of decoding calculation can be reduced as compared with the conventional method of performing scale-up and scale-down using only the variable scale amount.

Further, according to the seventh digital signal decoding method of the present invention, in the first digital signal decoding method, the representative value of the spectrum data of each unit is searched and the representative value is obtained. Based on this, a method of scaling up and down by a predetermined fixed scale amount and a scale up and down by a variable scale amount according to the data to be processed so that overflow does not occur in the process of inverse quantization and inverse orthogonal transformation. It is a method of selectively using a method of scaling down and a method of not performing scaling up and scaling down.

An eighth digital signal decoding method of the present invention is the same as the seventh digital signal decoding method, wherein the scale-up and scale-down are
This is a method of searching the maximum value of the spectrum data representative value of each unit and performing it based on the maximum value.

According to the seventh and eighth digital signal decoding methods, scale-up and scale-down are performed by a predetermined fixed scale amount based on the representative value of the spectrum data of each unit, or By selecting whether to scale up and down with a variable scale amount according to the calculation data or not to perform scale up and down, scale up and down can be performed using only a fixed amount of the conventional. Compared to the method
It is possible to reduce the amount of decoding calculation and improve the calculation accuracy. Further, the amount of decoding operation can be reduced as compared with the conventional method of performing scale-up and scale-down using only a variable scale amount.

[0165]

As described above, the digital signal decoding method according to the first aspect of the present invention is divided into a plurality of units according to the time range and the frequency domain, and the index value set for each unit. In the method of decoding a digital signal for decoding the spectrum data coded for each unit based on, the first step of searching a representative value from the index values in each unit, and based on this representative value And a second step of determining whether or not to scale the operand in the decoding operation, and a third step of scaling the operand based on the determination result in the second step. It is a method including and.

As a result, according to the above method, it is possible to avoid an arithmetic error due to overflow by scaling and to decode the data having a minute amplitude with high accuracy. Furthermore, since it is determined whether or not to perform scaling, scaling is not performed on data that does not need to be scaled. Therefore, the amount of calculation in the decoding process can be reduced. Further, since this judgment is made based on the representative value of the index value in each unit, the processing amount for this judgment is very small. As described above, according to the above method, it is possible to improve the accuracy in decoding a digital signal without increasing the amount of calculation.

According to a third aspect of the present invention, in addition to the method according to the first or second aspect , the digital signal has a scale amount for scaling in the third step, based on the representative value. It is a method including a fourth step of determining.

According to the above method, all scale amounts of the arithmetic processing are determined based on the representative value. Therefore, the scale amount can be determined by a very simple process. Therefore, according to this method, in addition to the effect of claim 1, the method of decoding a digital signal according to claim 1 can be easily realized, and the amount of calculation can be further reduced. Play.

According to a fourth aspect of the present invention, in addition to the method according to the first or second aspect , the digital signal decoding method further comprises a scaling amount for scaling in the third step. It is a method including a fifth step of making a determination based on a value to be calculated in the calculation.

According to the above method, the scale amount in the arithmetic processing is determined based on the operated value which is the input data of each arithmetic processing. Therefore, in each calculation process, optimum scaling can be applied to the calculated value. Therefore, according to this method, in addition to the effect of claim 1, there is an effect that the accuracy of decoding a digital signal can be made extremely high.

In addition to the method according to claim 1 or 2 , the digital signal decoding method according to claim 5 is characterized in that the scale amount for scaling in the third step is set to the representative value. It is a method including a sixth step of determining whether to determine based on the result of the determination, and determining based on the result of the determination.

According to the above method, the method of determining the scale amount is determined according to the spectrum data to be decoded, so that optimum scaling can be performed according to the characteristics of the spectrum data. . Furthermore, since this judgment is made using the representative value, the amount of calculation processing for this judgment is very small. Therefore, according to the above method, in addition to the effect of the first aspect, it is possible to perform scaling with an optimum scale amount without increasing the calculation amount.

A digital signal decoding method according to a sixth aspect of the present invention is the digital signal decoding method according to any one of the first to fifth aspects, wherein the index value is a spectrum in each unit. This is a method in which the representative value is the maximum value among the index values of all units, as well as the maximum amplitude value of the data.

According to the above configuration, the representative value becomes the maximum value in the amplitude value of the spectrum data. Therefore, since the scaling is performed based on this representative value, it is possible to reliably avoid the overflow in the arithmetic processing during the scaling. Further, in this method, since the scaling value and the maximum value of the index values only have to be selected in order to select the index value and the representative value, the selection process can be greatly simplified. As described above, according to the above method, in addition to the effects of the first to fourth aspects, it is possible to reliably avoid the overflow in the arithmetic processing without increasing the arithmetic amount.

According to a seventh aspect of the present invention, there is provided a method for decoding a digital signal, wherein the digital signal is divided into a plurality of units according to a time range and a frequency domain, and each unit is divided based on an index value set for each unit. A method of decoding a digital signal for decoding spectrum data quantized in, comprising: a seventh step of searching a representative value from index values in each unit; The eighth step of scaling the value and the scale amount for scaling in the eighth step are determined based on the representative value or the calculated value in the calculation. Judge,
A ninth step of determining the scale amount based on the determination result.

According to the above method, it is possible to avoid arithmetic error due to overflow by scaling, and also it is possible to accurately decode data of minute amplitude. Furthermore, since the method of determining the scale amount is determined according to the spectrum data to be decoded, optimal scaling can be performed according to the characteristics of the spectrum data. Furthermore, since this judgment is made using the representative value, the amount of calculation processing for this judgment is very small. Thus, according to the above method, it is possible to decode a digital signal by optimal scaling without increasing the amount of calculation.

The digital signal decoding method according to claim 9 of the present invention is the method according to claim 7 or 8 , wherein the index value is the maximum amplitude value of the spectrum data in each unit. In addition, the above-mentioned representative value is a method in which the index value of all units is the maximum value.

According to the above method, the representative value becomes the maximum value in the amplitude value of the spectrum data. Therefore, since the scaling is performed based on this representative value, it is possible to reliably avoid the overflow in the arithmetic processing during the scaling. Further, in this method, since the scaling value and the maximum value of the index values only have to be selected in order to select the index value and the representative value, the selection process can be greatly simplified. Thus, according to the above method, in addition to the effect of claim 6, it is possible to reliably avoid the overflow in the arithmetic processing without increasing the arithmetic amount. In addition, according to claim 12 of the present invention,
The digital signal decoding method is divided into the time domain and the frequency domain.
Therefore, it is divided into multiple units and set for each unit.
Is encoded for each unit based on the index value
Of the digital signal for decoding the spectrum data
In decoding method, IM for spectrum data
The process of performing DCT processing and the maximum of each block
Find the unit with the scale factor and
Block scale factor as a representative value of the block.
IMDCT treatment before the step of obtaining and IMDCT treatment
Before, in blocks divided by frequency band,
Based on the maximum scale factor of each unit,
In the process of calculation processing in IMDCT processing
Scales the operands so that there are no rows
Process and spectrum data processed by IMDCT
And a step of performing a reverse floating process to correct the amplitude value of
In the step of performing the above-mentioned reverse floating process,
Find the sum of the scale amounts used for scaling,
The value obtained by multiplying the obtained total by 3 is used in the IMDCT processing step.
Scale factor in each unit before processing
Subtract from the offset value of the index of the new off
Find the set value and set this new offset value from the reference value
Subtract and set a new offset value
The scale factor according to the new offset value.
The spectrum that has undergone IMDCT processing
This is a method of multiplying the amplitude value of data. According to the above method
Then, by scaling, in IMDCT processing
Avoids calculation errors due to overflow in calculation processing
And to decode very small amplitude data with high accuracy.
There is an effect that can be.

[Brief description of drawings]

FIG. 1 is a flowchart showing an operation flow of an IMDCT section in a mini disk recording / reproducing apparatus shown in a third embodiment of the present invention.

FIG. 2 is an explanatory diagram showing the configuration of the mini-disc recording / reproducing apparatus shown in the first embodiment of the present invention.

3 is an explanatory diagram showing a configuration of an audio compression circuit in the mini disk recording / reproducing device shown in FIG.

4 is an explanatory diagram showing an example of spectrum data unitized by the audio compression circuit shown in FIG. 3. FIG.

5 is an explanatory diagram showing a compression format of audio data output from the audio compression circuit shown in FIG.

6 is an explanatory diagram showing a configuration of an audio decompression circuit in the mini disk recording / reproducing device shown in FIG.

7 is an IMDC in the voice decompression circuit shown in FIG.
It is explanatory drawing which shows the structure of T section.

FIG. 8 is an explanatory diagram showing a configuration of an IMDCT section in the audio decompression circuit of the mini-disc recording / reproducing apparatus according to the second embodiment of the present invention.

FIG. 9 is an explanatory diagram showing a configuration of an IMDCT unit in an audio decompression circuit of a mini disk recording / reproducing device according to a third embodiment of the present invention.

FIG. 10 is an explanatory diagram showing a configuration of a voice decompression circuit configured not to perform scaling.

11 is an IMD in the audio decompression circuit shown in FIG.
It is explanatory drawing which shows the structure of CT part.

[Explanation of symbols]

1 Mini disk recording / reproducing device 6 Audio compression circuit 9 Voice expansion circuit 41 WL Development Department 42 SF development department 43 AS Development Department 44 SF maximum value determination unit 45 SF offset attaching section 46 inverse quantizer 47 IMDCT section 48 Reverse floating part 49 Window 50 band synthesis filter section 51 U (k) calculator 52 FFT calculator 53 u (n) calculator 54 y (n) calculator 55 Fixed scaling unit 56 Fixed scale amount loading section 57 Fixed scale amount memory 61 IMDCT section 62 Variable scaling unit 63 Variable scale amount determination unit 64 variable scale memory 71 IMDCT section 72 Scaling section

─────────────────────────────────────────────────── --Continued front page (56) Reference JP-A-6-318875 (JP, A) JP-A-4-302540 (JP, A) JP-A-6-164414 (JP, A) JP-A-8- 223574 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) H03M 7/30

Claims (11)

(57) [Claims] 1. Decoding of a digital signal for decoding spectrum data which is divided into a plurality of units according to a time range and a frequency domain and is coded for each unit based on an index value set for each unit. A first step of retrieving a representative value from the index values in each unit, and a first step of determining whether or not to scale the operand value in the operation for decoding based on the representative value. 2. A method of decoding a digital signal, comprising: the second step; and a third step of scaling the calculated value based on the judgment result of the second step. 2. IMDCT processing for spectrum data
Further including an IMDCT processing step of applying The third step is Prior to the IMDCT process, the frequency bands were divided.
The maximum scale factor of each unit in the block
Calculation processing in the IMDCT processing based on a large value
In order to prevent overflow in the process of
A method comprising the step of performing scaling.
1. A method of decoding a digital signal according to 1. 3. A fourth method for determining a scale amount for scaling in the third step based on the representative value.
3. The method of decoding a digital signal according to claim 1, further comprising the step of. 4. The method according to claim 1, further comprising a fifth step of determining a scale amount for scaling in the third step, based on an operand value in the operation.
Alternatively, the decoding method of the digital signal according to the item 2 . 5. A scale amount for scaling in the third step is determined based on the representative value, or
Alternatively, it is determined whether the decision on the basis of the calculation value in the calculation, based on the determination result, according to claim 1, characterized in that it comprises a sixth step of determining the amount of scale or
Or a method for decoding a digital signal according to item 2 . Wherein the index value, with the largest amplitude value of the spectrum data of each unit, said representations, according to claim 1 to 5, characterized in that the maximum value of the index values of all units A method for decoding a digital signal according to any one of the above. 7. A digital signal decoding for decoding spectrum data, which is divided into a plurality of units according to a time range and a frequency domain and is coded for each unit based on an index value set for each unit. In the coding method, a seventh step of searching a representative value from the index values in each unit, an eighth step of scaling the operand value in the calculation for decoding, and a scaling step in the eighth step. A ninth step of determining whether the scale amount for determining based on the representative value or the calculated value in the calculation and determining the scale amount based on the determination result. A method for decoding a digital signal, comprising: 8. IMDCT processing for spectrum data
Further including the step of applying The eighth step is Prior to the IMDCT process, the frequency bands were divided.
The maximum scale factor of each unit in the block
Calculation processing in the IMDCT processing based on a large value
In order to prevent overflow in the process of
A method comprising the step of performing scaling.
7. The method for decoding a digital signal according to 7. 9. The index value, with the largest amplitude value of the spectrum data of each unit, said representations, to claim 7 or 8, characterized in that the maximum value of the index values of all units A method for decoding a digital signal as described. 10. IMDCT for spectrum data
The process of applying the treatment, Amplitude value of spectrum data processed by IMDCT
And a step of performing a reverse floating process for correcting
The digit according to any one of claims 1 to 9, characterized by
Tal signal decoding method. 11. In the process of performing the reverse floating process
Is Calculate the sum of the scale amounts used for the above scaling
Because The value obtained by multiplying the obtained total by 3 is obtained before the IMDCT processing,
Of the scale factor index in each unit
Subtract from the offset value to obtain a new offset value, Subtract this new offset value from the reference value to
To a new offset value, Scale factor according to this new offset value
Is the scale value, and the IMDCT processed spectrum
Claim to multiply the amplitude value of tram data
Item 10. A method for decoding a digital signal according to Item 10.