JP3889738B2 - Inverse quantization apparatus, audio decoding apparatus, image decoding apparatus, inverse quantization method, and inverse quantization program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a reverse quantization device with small circuit scale and small operation error. <P>SOLUTION: Reverse quantization processing for inputting a scale factor index Scf, bit allocation Ba and a quantization code X and outputting a reverse quantization value Y is performed by reducing frequency of multiplication by using bit shift processing. Even when the multiplication is performed only once, an error to be generated in a single multiplication is suppressed to be small not only by taking out a part of integer power of 2 as first shift quantity S1 from the scale factor index but taking out a part of the integer power of 2 as second shift quantity S2 from the bit allocation Ba and a reference value R and replacing multiplication processing to the quantization code X with the bit shift processing as much as possible. <P>COPYRIGHT: (C)2005,JPO&amp;NCIPI

Description

  The present invention relates to an inverse quantization device, an audio decoding device, an image decoding device, an inverse quantization method, and an inverse quantization program, and more specifically, an inverse from an input value X, a scale factor index Scf, and a bit allocation Ba. The present invention relates to an inverse quantization device, an audio decoding device, an image decoding device, an inverse quantization method, and an inverse quantization program that generate a quantized value Y.

  In order to handle an analog signal by digital signal processing, in addition to sampling for discretizing the domain of the signal, a quantization process for discretizing the range of the signal is necessary. In addition, when a digital signal that has already been discretized in a range is handled with less data, the signal may be further coarsened. This process is also called quantization.

  In this quantization, a value after discretization is called a quantized value, and a code assigned with the quantized value is called a quantized code. The interval between adjacent quantized values is called a quantization step width, and the number of quantized values included in the entire range is called a quantization step number.

  Conversely, when digital signal processing is performed and then output as an analog signal, the process of restoring the original continuous value from the quantized code or the value discretized with the original small step width is required. This process is called inverse quantization.

  In the process of inverse quantization, conversion is performed in which the quantization value corresponding to the quantization code is converted to a value after inverse quantization (inverse quantization value).

  The quantization value approximates a continuous value in the quantization process. For this reason, an error (quantization error) from the original value occurs in the quantized value.

  Accordingly, the value before quantization cannot be restored in the inverse quantization process, and the restored value is an approximate value including a quantization error.

  In addition, in the quantization process, there is a case where scaling processing is performed in which the signal is normalized after being in a range (dynamic range) taken by the actual signal. When performing the scaling process, the normalization coefficient (scale factor) and the quantization code for the normalized signal are individually held. When scaling processing is performed, in the process of inverse quantization, inverse quantization is performed on the quantized code for the signal after normalization, and then inverse normalization is performed using a scale factor to calculate an inverse quantization value. Processing is performed.

  Since the quantization code can be assigned only to the range where the signal actually exists by the scaling process, even if the number of quantization steps is limited, the quantization error is suppressed and the input signal is Effective quantization is possible.

  The quantization and scaling techniques as described above are used in a wide range of fields such as image coding and audio coding represented by the MPEG (Motion Picture Experts Group) standard.

  General explanations regarding quantization are described in, for example, Chapter 6 of “Digital Signal Processing of Images, Nikkan Kogyo”, Chapter 1 of “Sound Systems and Digital Processing, Institute of Electronics, Information and Communication Engineers”, and the like. Yes.

  Here, specific dequantization is taken by taking as an example the quantization method employed in the MPEG1-Audio Layer II method described in ISO / IEC11172-3, which is one of standard audio encodings, A conventional apparatus that performs inverse scaling processing will be described.

  The MPEG1-Audio Layer II system described in ISO / IEC 11172-3 is an audio encoding system for the purpose of compressing the data amount of an audio signal while maintaining high sound quality, and hardly deteriorates the sound quality. It is possible to compress the data amount of the audio signal to about one fifth.

  When data is stored or transmitted, data compression is performed by an encoding process that converts time-series audio signals into MPEG1-Audio Layer II format data.

  When data is reproduced, MPEG1-Audio Layer II format data compressed by decoding processing is converted into a time-series audio signal. In this decoding process, when the amplitude of the audio signal is encoded, inverse quantization and inverse scaling process as described later are performed.

  Details of the decoding apparatus that performs the above-described decoding processing are described in ISO / IEC 11172-3.

  Next, the inverse quantization and inverse scaling processing employed in the MPEG1-Audio Layer II system described in ISO / IEC 11172-3 will be described.

  The input values in the inverse quantization and inverse scaling processes are a scale factor index Scf that is a code corresponding to the coefficient of the scaling process, a number of quantization steps, a bit allocation Ba that is a code corresponding to the quantization step width, and an audio signal. Quantized code X, which is a code corresponding to the amplitude of.

  The scale factor index Scf, the bit allocation Ba, and the quantization code X are stored in a predetermined format in the data of the MPEG1-Audio Layer II format, and a bit string of a corresponding portion is extracted from this data and subjected to predetermined processing. Generated.

  Hereinafter, this point will be described in detail.

[Retrieving numbers from a stream]
The scale factor index Scf is stored as a 6-bit bit string in the MPEG1-Audio Layer II format. This bit string is read as an integer value as it is and used as the scale factor index Scf. However, in the ISO / IEC 11172-3, it is prohibited that all of these 6 bits become “1”, so that the scale factor index Scf to be read is an integer value in the range of 0 to 62. Become.

  Next, generation of the bit allocation Ba will be described. First, data that is the basis of the bit allocation Ba is stored in a predetermined number of bits in the MPEG1-Audio Layer II format. Here, the original data is referred to as index data. The index data is an integer value corresponding to a predetermined number of bits. The number of bits of the index data is the table B.1 of ISO / IEC11172-3. 2 to 4 bits as defined in 2.

  A method for converting index data into the number of steps is also described in Table B. of ISO / IEC 11172-3. 2 is stipulated. As will be described next with reference to FIG. 11, the number of steps is determined according to the sampling frequency, the bit rate per channel, and the frequency band (subband) to which the audio signal to be dequantized belongs.

  FIG. 11 shows the table B. of ISO / IEC 11172-3. It is the figure which showed an example of 2.

  Table B. of ISO / IEC 11172-3. 2 has a plurality of tables, and which table is used is determined based on the sampling frequency and the bit rate per channel.

  Referring to FIG. 11, sb represents a subband. One column under the word nbal is a value indicating how many bits the index data is included in the MPEG1-Audio Layer II format. On the right side of nbal, the number of steps corresponding to each index data 0 to 15 is described. Here, the number of steps is described in “Number of steps” in ISO / IEC 11172-3.

  A table is selected according to the sampling frequency and the bit rate per channel, and the number of bits of the index data corresponding to the frequency band (subband) to which the audio signal to be inversely quantized belongs is obtained by the table. . Then, the index data is read out from the stream, and again, using this table, the number of steps corresponding to the index data 0 to 15 in the subband of the audio signal is acquired. The acquired number of steps is a value corresponding to the number of quantization steps.

  As shown in FIG. 11, the types of steps that can be taken in the MPEG1-Audio Layer II system are 3, 5, 7, 9, 15, 31, 63, 127, 255, 511, 1023, 2047, 4095, 8191. , 16383, 32767, 65535. A different code is assigned to each of the 17 types of steps, and the code thus obtained is defined as a bit allocation Ba.

  FIG. 12 is a diagram showing the correspondence between the bit allocation Ba and the number of steps.

  In FIG. 12, integers from 0 to 16 are assigned as bit allocation Ba in order from the smallest number of steps.

  The quantized code X is generated based on a numerical value stored in MPEG1-Audio Layer II format data with a predetermined number of bits. This number of bits corresponds to the table B.1 of ISO / IEC11172-3 illustrated in FIG. 4 and FIG.

  FIG. 13 shows the table B. of ISO / IEC 11172-3. FIG.

  With reference to the number of steps, the number of samples per codeword (Samples per codeword), and the number of bits per codeword (Bits per codeword) shown in FIGS. Thus, a predetermined number of bits is determined. Samples per codeword is a value indicating how many audio sample data are encoded in one code. Bits per codeword is a value indicating how many bits the code is encoded.

  In ISO / IEC11172-3, one audio sample is encoded with one code and three audio samples are encoded with one code by a combination of Samples per codeword and Bits per codeword. There are cases. However, in any case, an integer value between 0 and the number of steps is acquired for each audio sample by the process defined in ISO / IEC 11172-3. The integer value obtained from MPEG1-Audio Layer II format data is a fixed-point format small number in which the most significant bit of the number of steps when the step number is expressed in binary is the sign bit and the subsequent bits are the decimal part. It is stored as a numerical value, and further converted into a format in which its sign bit is inverted.

For example, a case will be described in which the most significant bit is held as a sign bit on a 24-bit register such as a DSP (digital signal processor). It is assumed that the bit allocation Ba = 5 and the integer value obtained from the data = 5 (101 in binary representation). From FIG. 12, since the number of steps is 31 if bit allocation Ba = 5, the number of digits when the acquired integer value is expressed in binary is 5. This is because the 5-digit binary number is 31 at the maximum. In this case, the value is packed on the most significant bit side so that the fifth digit is the most significant bit on the DSP register. The stored value is in binary notation,
0010 1000 0000 0000 0000 0000 is held.

Furthermore, since the sign bit is inverted, the value to be held is represented by a binary number,
1010 1000 0000 0000 0000 0000

  Therefore, if this binary number is read in a 24-bit fixed-point format, the finally held value is -0.6875. The decimal value obtained in this way is defined as a quantization code X.

The following equation (11) is a conversion equation for inverse quantization and inverse scaling at the time of decoding of the MPEG1-Audio Layer II system described in ISO / IEC 11172-3.
Y = 2 ^{(P1 [Scf] / M)} * P2 [Ba] * (X + P3 [Ba]) (11)
Here, “*” is an operation code representing multiplication, and “/” is an operation code representing division. P2 [Ba] and P3 [Ba] are decimal values determined with respect to the bit allocation Ba, and are defined as values C and D in the table shown in FIG. The association between the bit allocation Ba and the decimal values P2 [Ba] and P3 [Ba] is determined by referring to FIGS.

  The above processing is performed from the data of the MPEG1-Audio Layer II format, the scale factor index Scf, the bit allocation Ba, and the quantization code X are taken out and converted into the inverse quantized value Y according to the equation (11). Quantization and inverse scaling processing are performed.

[Conventional inverse quantization and inverse scaling]
Next, a conventional inverse quantization and inverse scaling apparatus that implements inverse quantization and inverse scaling processing employed in the MPEG1-Audio Layer II system described in ISO / IEC 11172-3 will be described.

  FIG. 14 is a schematic block diagram for explaining a conventional inverse quantization and inverse scaling apparatus.

  Referring to FIG. 14, scale factor index Scf, bit allocation Ba, and quantization code X are input to this apparatus. These input values are generated based on values read out from the data in the MPEG1-Audio Layer II format as described above. The table reference unit 110 receives the scale factor index Scf, refers to the conversion table To1, and generates and outputs the scaling coefficient Sv.

The conversion table To1 stores a scaling coefficient Sv that is a decimal value corresponding to the scale factor index Scf in the range of 0 to 62. The scaling coefficient Sv is a value obtained as a result of calculation in advance by the following equation (12).
Sv = 2 ^{((3-Scf) / 3)} (12)
The table reference unit 111 receives the bit allocation Ba as an input, refers to the conversion table To2, and generates and outputs conversion coefficients Co1 and Co2. In the conversion table To2, the C of FIG. 13 corresponding to the bit allocation Ba of 0 to 16 shown in FIG. 12 is stored as the value of the conversion coefficient Co1, and the corresponding D of FIG. 13 is the conversion coefficient Co2. Is stored as the value of.

The conversion unit 112 receives the conversion coefficients Co1 and Co2 and the quantization code X as inputs, and generates and outputs a conversion value Zo. The conversion in the conversion unit 112 is performed by the addition and multiplication processing shown in the following equation (13).
Zo = Co1 * (X + Co2) (13)
The conversion unit 113 receives the scaling coefficient Sv and the conversion value Zo as inputs, and generates and outputs an inverse quantization value Y. The conversion in the conversion unit 113 is performed by a multiplication process represented by the following equation (14).
Y = Sv * Zo (14)
As described above, conventionally, inverse quantization and inverse scaling have been performed.

  FIG. 15 is a flowchart in the case where conventional inverse quantization and inverse scaling processing are performed using a computer such as a DSP.

  Referring to FIG. 15, in step S101, inverse quantization and inverse scaling processing are started. Subsequently, a table To1 reference process is performed in step S102. In the DSP, the scale factor index Scf, the bit allocation Ba, and the quantization code X are read out from the MPEG1-Audio Layer II format data through a predetermined process and stored in a random access memory (RAM). And

  In the table To1 reference processing, the scale factor index Scf is read from the RAM, the scaling coefficient Sv is generated by referring to the conversion table To1 stored in the read only memory (ROM), and stored in the RAM. The conversion table To1 is the same as that in FIG.

  When step S102 ends, table To2 reference processing is performed in step S103. In step S103, the bit allocation Ba is read from the RAM, the conversion coefficients Co1 and Co2 are generated by referring to the conversion table To2 stored in the ROM, and stored in the RAM. The conversion table To2 is the same as that in FIG.

Subsequently, in step S104, a first conversion process is performed. In step S104, the conversion coefficients Co1 and Co2 and the quantization code X are read from the RAM, the addition and multiplication processing shown in the following equation (15) is performed, and a conversion value Zo is generated and stored in the RAM.
Zo = Co1 * (X + Co2) (15)
When step S104 ends, a second conversion process is subsequently performed in step S105. In step S105, the scaling coefficient Sv and the conversion value Zo are read from the RAM, the multiplication processing shown in the following equation (16) is performed, and the inverse quantization value Y is generated and stored in the RAM.
Y = Sv * Zo (16)
Finally, in step S106, the inverse quantization and inverse scaling processing ends.

  Even when the DSP is used as described above, inverse quantization and inverse scaling processing are performed.

Note that the above description of the prior art is not limited to the MPEG1-Audio Layer II system described in ISO / IEC 11172-3, but is the one that performs conversion according to Equation (11) among the inverse quantization and inverse scaling processes. It can be applied without losing generality.
Y = 2 ^{(P1 [Scf] / M)} * P2 [Ba] * (X + P3 [Ba]) (11) re-displayed A related technique for reducing the operation load of inverse quantization processing is disclosed in Japanese Patent Laid-Open No. 8-292975. (Patent Document 1).
Japanese Patent Application Laid-Open No. 8-292895

  When performing inverse quantization and inverse scaling according to Equation (11), the conventional method described with reference to FIG. 14 needs to perform two multiplications for one audio sample.

  In general, when the multiplication is realized as hardware, a large-scale circuit is required, and when it is realized as software, a large number of processing steps are required. Therefore, when this inverse quantization and inverse scaling are realized by a DSP or the like, there arises a problem that the cost becomes high.

  Further, when digital signal processing is realized using a DSP or the like, since the number of digits of calculation is limited, errors are accumulated by repeating multiplication. In addition, when calculating in the fixed-point format, if the first multiplication is multiplied by a small value and the second multiplication is multiplied by a large value, the first multiplication reduces the number of significant digits, and the second multiplication. The phenomenon that the error is amplified occurs. Thus, there arises a problem that a larger calculation error is more likely to occur as the number of multiplications increases. There is also a problem that the error is amplified if the order of multiplication is not taken into consideration.

  The present invention has been proposed to solve such a problem, and provides an inverse quantization device and an audio decoding device that can accurately realize inverse quantization and inverse scaling with a small number of multiplications. Objective.

  In summary, the present invention is an inverse quantization apparatus that receives a quantization code X, a scale factor index Scf, and a bit allocation Ba, performs inverse quantization and inverse scaling processing, and outputs an inverse quantization value Y. Storage means, first reference means, second storage means, second reference means, conversion means, and shift processing means.

The first storage means stores a shift amount S1 and a reference value R corresponding to each scale factor index Scf calculated in advance based on the following expressions (1) and (2). The first reference means refers to the first storage means to obtain the shift amount S1 and the reference value R corresponding to the input scale factor index Scf. The second storage means stores the reference value R calculated in advance based on the following expressions (3), (4), and (5) and the shift amount S2, the conversion coefficients C1 and C2 corresponding to each bit allocation Ba. . The second reference means refers to the second storage means to obtain the shift value S2 corresponding to the reference value R output from the first reference means and the input bit allocation Ba, and the transform coefficients C1 and C2. . The conversion means performs conversion according to the following equation (6) using the input quantization code X and the conversion coefficients C1 and C2 output from the second reference means, and outputs a conversion value Z. The shift processing means performs a shift process according to the following equation (7) using the shift amounts S1 and S2 respectively obtained by the first and second reference means for the conversion value Z output from the conversion means and performs the reverse operation. The quantized value Y is output. Shift processing means.
S1 = int (P1 [Scf] / M) (1)
R = mod (P1 [Scf], M) (2)
S2 = int ((R / M) + log ₂ (P2 [Ba]) (3)
C1 = (2 ^{(R / M)} * P2 [Ba]) * (2 ^(-S2) ) (4)
C2 = P3 [Ba] (5)
Z = C1 * (X + C2) (6)
Y = Z * 2 ^{(S1 + S2)} (7)
However, P1 [Scf] indicates a predetermined integer for the value of each scale factor index Scf, and P2 [Ba] and P3 [Ba] indicate a predetermined decimal for the value of each bit allocation Ba. M represents a predetermined positive integer.

  Preferably, the shift processing means performs the shift process by performing a first shift operation corresponding to the shift amount S1 and further performing a second shift operation corresponding to the shift amount S2.

  Preferably, the shift processing means performs the shift processing by collectively performing a shift operation corresponding to the total value of the shift amount S1 and the shift amount S2.

Preferably, the transforming means is based on transform coefficients C1 and C2 and a quantization code X, and the following formula Z = C1 * X + C2
The conversion value Z is generated according to

  Preferably, P1 [Scf] is 3-Scf and M is 3.

  Or it is an audio decoding apparatus which performs a decoding process based on the audio encoding system including a dequantization process at the time of decoding, Comprising: One of the said dequantization apparatuses is provided.

  Or it is an image decoding apparatus which performs a decoding process based on the image coding system including a dequantization process at the time of decoding, Comprising: One of the said dequantization apparatuses is provided.

According to another aspect of the present invention, there is provided an inverse quantization method that receives a quantization code X, a scale factor index Scf, and a bit allocation Ba, performs inverse quantization and inverse scaling processing, and outputs an inverse quantization value Y. Referring to the first storage means for storing the shift amount S1 and the reference value R corresponding to each scale factor index Scf calculated in advance based on the following equations (1) and (2), the inputted scale factor index Scf A step of obtaining a shift amount S1 and a reference value R corresponding to each of the reference values R, and a shift amount S2 corresponding to each reference value R and each bit allocation Ba calculated in advance based on the following equations (3), (4), (5) Referring to the second storage means for storing the conversion coefficients C1 and C2, the obtained reference value R and the shift amount S corresponding to the input bit allocation Ba , Obtaining transform coefficients C1 and C2, using the input quantization code X and the obtained transform coefficients C1 and C2, performing transformation according to the following equation (6), and outputting a transformed value Z: And a step of performing a shift process according to the following equation (7) using the shift amounts S1 and S2 obtained for the converted value Z and outputting an inverse quantized value Y.
S1 = int (P1 [Scf] / M) (1)
R = mod (P1 [Scf], M) (2)
S2 = int ((R / M) + log ₂ (P2 [Ba]) (3)
C1 = (2 ^{(R / M)} * P2 [Ba]) * (2 ^(-S2) ) (4)
C2 = P3 [Ba] (5)
Z = C1 * (X + C2) (6)
Y = Z * 2 ^{(S1 + S2)} (7)
However, P1 [Scf] indicates an integer predetermined for each scale factor index Scf, and P2 [Ba] and P3 [Ba] indicate a predetermined decimal number for each bit allocation Ba value. M represents a predetermined positive integer.

  Or it is a dequantization program and makes a computer perform the process based on the said dequantization method.

  According to the present invention, the inverse quantization and inverse scaling device performs multiplication only once. Therefore, when implemented as hardware, the circuit scale can be reduced. When implemented as software, the number of processing steps can be reduced. This makes it possible to perform inverse quantization and inverse scaling processing at a low cost.

  In addition, since the multiplication is performed only once, the calculation error is not accumulated, and the error due to the previous multiplication is not amplified by the subsequent multiplication. Therefore, more accurate inverse quantization and inverse scaling processing can be performed.

  Further, according to the present invention, it is possible to realize an audio decoding apparatus that performs accurate decoding with a small number of multiplications for an audio encoding system including inverse quantization and inverse scaling processing.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the drawings, the same reference numerals indicate the same or corresponding parts.

[Embodiment 1]
FIG. 1 is a block diagram showing a configuration example of the audio decoding apparatus according to the first embodiment.

  Referring to FIG. 1, audio decoding apparatus 1 includes a format analysis unit 2, an inverse quantization / inverse scaling unit 3, and a filter unit 4.

  The format analysis unit 2 receives an MPEG stream for one frame (MPEG1-Audio Layer II data stream) as an input, separates various information stored in the frame according to a format defined by the standard, Take out in the prescribed format. This extracted information includes the scale fuck index Scf, the bit allocation Ba, and the quantization code X.

  The inverse quantization / inverse scaling unit 3 receives the separated scale factor index Scf, bit allocation Ba, and quantization code X from the format analysis unit 2, performs inverse quantization and inverse scaling processing, and performs an inverse quantization value Y is generated.

  In the MPEG1-Audio Layer II format, audio data is stored after being subjected to some kind of frequency conversion, and the inverse quantized value Y obtained here is also a value in the frequency domain.

  The filter unit 4 performs a filter process on the inverse quantized value Y to perform conversion to the time domain, and generates and outputs audio data ADATA that is time-series data.

  Through the above process, one frame of MPEG1-Audio Layer II format data is decoded. By repeating this for each frame, the MPEG1-Audio Layer II decoding process is performed.

  FIG. 2 is a block diagram showing the configuration of the inverse quantization / inverse scaling unit 3 in FIG.

  Referring to FIG. 2, the scale factor index Scf, the bit allocation Ba, and the quantization code X are inputs to the inverse quantization / inverse scaling unit 3, and these values are obtained from the MPEG1-Audio Layer II format data. This is a value extracted by the format analysis unit 2 in FIG.

  The inverse quantization / inverse scaling unit 3 includes a table reference unit 10, a conversion table T 1, a table reference unit 11, a conversion table T 2, a conversion unit 12, and a shift processing unit 13.

The table reference unit 10 receives the scale factor index Scf as an input, generates a shift amount S1 and a reference value R with reference to the conversion table T1, and outputs them. The conversion table T1 stores a shift amount S1 and a reference value R respectively corresponding to the scale factor index Scf that is an integer value in the range of 0 to 62. The shift amount S1 is a value obtained in advance by the following equation (17). The reference value R is a value obtained in advance by the following equation (18).
S1 = int ((3-Scf) / 3) (17)
R = mod (3-Scf, 3) (18)
Here, int (A) is an operation for generating a maximum integer that does not exceed A. Mod (A, B) is an operation for calculating A-int (A / B) * B for integer A and integer B. At this time, the reference value R is an integer value in the range of 0-2.

The table reference unit 11 receives the bit allocation Ba and the reference value R as inputs. Then, the table reference unit 11 refers to the conversion table T2 to generate and output the shift amount S2 and the conversion coefficients C1 and C2. The conversion table T2 stores a shift amount S2 and conversion coefficients C1 and C2. The shift amount S1 is an integer value calculated in advance by the following equation (19). The conversion coefficient C1 is a decimal value calculated in advance by the following equation (20). The conversion coefficient C2 is a decimal value predetermined by the following equation (21).
S2 = int ((R / 3) + log ₂ (P2 [Ba]) (19)
C1 = (2 ^{(R / 3)} * P2 [Ba]) * (2 ^(-S2) ) (20)
C2 = P3 [Ba] (21)
Here, P2 [Ba] is a value obtained from the bit allocation Ba with reference to FIG. 13C and FIG. P3 [Ba] is a value obtained from the bit allocation Ba with reference to FIG. 13D and FIG.

The conversion unit 12 receives the conversion coefficients C1 and C2 and the quantization code X as inputs, and generates and outputs a conversion value Z. The conversion value Z is obtained by the conversion unit 12 by the addition and multiplication processing shown by the following equation (6).
Z = C1 * (X + C2) (6)
The shift processing unit 13 receives the shift amounts S1 and S2 and the converted value Z as inputs, and generates and outputs an inverse quantized value Y. The shift processing unit 13 calculates the inverse quantization value Y based on the calculations shown in the following formulas (22) and (23). Y is eventually the value represented by equation (7).
Z2 = Z * 2 ^(S1) (22)
Y = Z2 * 2 ^(S2) (23)
Y = Z * 2 ^{(S1 + S2)} (7)
At this time, multiplying by an integer power of 2 can be easily realized as a bit shift process.

By the following equation modification, the following equation (24) obtained by substituting P1 [Scf] = 3-Scf and M = 3 for equation (11) of the conventional method was obtained by equation (23). It is shown that the inverse quantized values Y are essentially equivalent.
Y = 2 ^{((3-Scf) / 3)} * P2 [Ba] * (X + P3 [Ba]) (24)
Y = Z2 * 2 ^(S2) (23) re-development If Z2 is expanded using equation (22) in equation (23),
Y = Z * 2 ^(S1) * 2 ^(S2) ... (25)
Furthermore, if Z is expanded using Equation (6),
Y = C1 * (X + C2) * 2 ^(S1) * 2 ^(S2) (26)
Further, when C1 and C2 are expanded using the equations (20) and (21), Y = (2 ^{(R / 3)} * P2 [Ba]) * 2 ^(-S2) * (X + P3 [Ba]) * 2 ^{(S1 ) * 2 (S2) ... (} 27)
Organize the formula
Y = (2 ^{(R / 3)} * P2 [Ba]) * (X + P3 [Ba]) * 2 ^(S1) (28)
Further, using equations (17) and (18), R and S1 are expanded,
Y = (2 ^{(mod (3-Scf, 3) / 3)} * P2 [Ba]) * (X + P3 [Ba]) * 2 ^{(int ((3-Scf) / 3))} (29)
Organize the formula
Y = (2 ^{(mod (3-Scf, 3) / 3 + int ((3-Scf) / 3))} * P2 [Ba]) * (X + P3 [Ba]) (30)
From the definitions of mod and int,
Y = 2 ^{((3-Scf) / 3)} * P2 [Ba] * (X + P3 [Ba]) (31)
Thus, equation (31) is the same as equation (24). Therefore, it was shown that the inverse quantized value Y obtained by Equation (23) is essentially equivalent to Equation (11) of the conventional method.

  According to the present invention, the inverse quantization and inverse scaling device performs multiplication only once. Therefore, when implemented as hardware, the circuit scale can be reduced. When implemented as software, the number of processing steps can be reduced. This makes it possible to perform inverse quantization and inverse scaling processing at a low cost.

  In addition, since the multiplication is performed only once, the calculation error is not accumulated, and the error due to the previous multiplication is not amplified by the subsequent multiplication. Therefore, more accurate inverse quantization and inverse scaling processing can be performed. Further, even when multiplication is performed only once, not only the integer power of 2 is extracted from the scale factor index as the first shift amount S1, but also the integer power of 2 from the bit allocation Ba and the reference value R. Is taken out as the second shift amount S2, and the multiplication process for the quantized code X is replaced with the bit shift process as much as possible to suppress the occurrence of errors by a single multiplication.

[Modification 1]
The shift processing unit 13 receives the shift amounts S1 and S2 and the conversion value Z as input, first obtains the total shift amount according to the following equation (32), and then calculates both the shift amounts S1 and S2 according to the equation (33). Bit shift processing may be performed collectively.
S3 = S1 + S2 (32)
Y = Z * 2 ^(S3) (33)
Thereby, inverse quantization and inverse scaling processing can be realized by one bit shift processing, and the processing may be further simplified.

[Modification 2]
The conversion table T2, instead of the equation (21) may be stored coefficients C 2 which is previously calculated based on the equation (34), the conversion unit 12 receives the coefficients C 2 from the table reference unit 11, the formula ( Instead of 6), addition and multiplication processing may be performed based on equation (35) to generate and output a converted value Z.
C2 = C1 * P3 [Ba] (34)
Z = C1 * X + C2 (35)
The coefficient C1 in the equations (34) and (35) is defined by the equation (20) as in the first embodiment. Even in such a modification, inverse quantization and inverse scaling processing can be realized in the same manner.

  In this embodiment, the case where P1 [Scf] = 3-Scf and M = 3 is described as an example, but the content of the present invention is not limited to this case.

Generally, P1 [Scf] is defined as an integer value with respect to the scale factor index Scf, and if M is an integer value, the values stored in the conversion table T1 and the conversion table T2 are expressed by the formula (17 ) To (21), instead of the values determined in (1) to (5), store the values determined using the values in (1) to (5) and perform the same processing to perform inverse quantization and inverse scaling processing. Can be realized.
S1 = int (P1 [Scf] / M) (1)
R = mod (P1 [Scf], M) (2)
S2 = int ((R / M) + log ₂ (P2 [Ba]) (3)
C1 = (2 ^{(R / M)} * P2 [Ba]) * (2 ^(-S2) ) (4)
C2 = P3 [Ba] (5)
In this embodiment, the values defined in ISO / IEC 11172-3 are used as the values of P2 [Ba] and P3 [Ba]. However, the contents of the present invention are not limited to this case. If it is a predetermined coefficient, the inverse quantization and the inverse scaling process can be realized by performing the same process.

  In the prior art and the embodiment, the case where the value read out from the data in the MPEG1-Audio Layer II format is packed as the quantization code X on the most significant bit side has been described. The digit position of the transform coefficient C2 stored in T2 is preliminarily matched with the digit position of the quantization code X packed to the least significant bit side, and the shift amount S2 stored in the conversion table T2 is offset by the number of digits. By adding in advance, inverse quantization and inverse scaling processing can be realized by the same processing.

  In the prior art and the embodiment, the case where the value read from the data of the MPEG1-Audio Layer II format is used as the quantization code X with the most significant bit inverted is used. Considering that the inversion corresponds to adding -1.0 in the two's complement representation of the fixed-point value, an offset of -1.0 minute is added in advance to the conversion coefficient C2 stored in the conversion table T2. By doing so, inverse quantization and inverse scaling processing can be realized by the same processing.

  As described above, the contents of the present invention are not limited to the case shown in the embodiment regarding the digit position for holding the value and the presence / absence of the sign bit inversion process, and the value is padded with the upper bits. In this case, the present invention can be similarly applied to the case where the lower bit is packed, the case where it is adjusted to another appropriate position, the case where the sign bit is inverted, and the case where there is no sign bit.

[Embodiment 2]
FIG. 3 is a block diagram illustrating a configuration example of the audio decoding device 21 according to the second embodiment.

  FIG. 4 is a flowchart for explaining a decoding process performed in the audio decoding apparatus shown in FIG.

  Referring to FIGS. 3 and 4, operation unit 23 is a block that reads and executes the program code in which the encoding process shown in FIG. 4 is described from ROM 24. The calculation unit 23 performs various calculations necessary for the decoding process and accesses the ROM 24, RAM 25, input I / F unit 22, and output I / F unit.

  The ROM 24 is a readable recording means, and is a block for storing program codes describing the decoding process of FIG. 4 and table data necessary for the decoding process.

  The RAM 25 is a recording means that can be read and written, and is a block that stores MPEG data used for decoding processing and temporarily generated calculation results.

  The input I / F unit is a block for acquiring the MPEG stream MS, and is a storage medium storing an MPEG stream such as a memory, a CD (compact disc), an HDD (hard disk drive), or an SPDIF (SONY / Philips Digital Interface Format). ), An interface for connecting to a communication medium for transmitting MPEG streams such as the Internet, and a buffer for temporarily storing the input MPEG streams.

  The output I / F unit 26 is a block for outputting audio data ADATA, a buffer for temporarily storing the decoded audio data, a playback device such as an amplifier and a speaker, and a recording device such as an MD and a CD It consists of an interface for connecting to.

  Note that the configuration shown in FIG. 3 is also used in the decoding process of image-encoded data represented by the MPEG standard. Therefore, the present invention can be applied not only to an audio decoding device but also to an image decoding device as long as it performs inverse quantization / inverse scaling processing.

  Next, processing performed by the audio decoding device 21 will be described.

  First, the decoding process is started in step S1 of FIG. In step S2, MPEG data is acquired. In step S <b> 2, the calculation unit 23 acquires MPEG data for one frame via the input I / F unit 22 and stores it in the RAM 25.

  Subsequently, format analysis is performed in step S3. In the format analysis, the MPEG data stored in the RAM 25 is read out, and various information stored in the frame is separated according to the format defined as the standard of the MPEG1-Audio Layer II system, and is taken out in a predetermined format. The extracted information is stored in the RAM 25.

  As an example of the extracted information, there are a scuff factor index Scf, a bit allocation Ba, a quantization code X, and the like which are input in the subsequent inverse quantization / inverse scaling process in step S4.

  In the inverse quantization / inverse scaling in step S4, the scale factor index Scf, the bit allocation Ba, and the quantization code X are read from the RAM 24, and the inverse quantization and the inverse scaling processing are performed using them, and the inverse quantization value Y Is generated. The inverse quantized value Y is stored in the RAM 25. The table data necessary for the inverse quantization / inverse scaling process is appropriately read from the RAM 24 and used.

  In the MPEG1-Audio Layer II format, audio data is stored in a state where a certain frequency conversion is performed, and the inverse quantization value Y obtained as a result of step S4 is also a value in the frequency domain. Therefore, in the subsequent filtering process in step S5, the inverse quantization value Y is read from the RAM 25, the filtering process is performed, and the conversion to the time domain is performed.

  As a result, audio data that is time-series data is generated and stored in the RAM 25. In step S6, audio data is output. That is, the generated audio data is read from the RAM 25 and the audio data is output via the output I / F unit 26.

  Through the above process, one frame of MPEG1-Audio Layer II format data is decoded. By repeating this process for each frame, the MPEG1-Audio Layer II system decoding process is performed.

  In step S7, the decoding process ends.

  FIG. 5 is a flowchart showing in more detail the inverse quantization / inverse scaling process of step S4 in FIG.

  Referring to FIG. 5, in step S11, inverse quantization / inverse scaling processing is started. The scale factor index Scf, the bit allocation Ba, and the quantization code X are inputs of this processing, and are read out from the data in the MPEG1-Audio Layer II format and stored in the RAM 25.

  First, in step S12, processing for referring to the table T1 is performed. In step S12, the scale factor index Scf is read from the RAM 25, the conversion table T1 stored in the ROM 24 is referred to, and the corresponding shift amount S1 and reference value R are read from the ROM. The read shift amount S1 and reference value R are stored in the RAM 25. Note that conversion table T1 is similar to conversion table T1 shown in FIG. 2, and therefore description thereof will not be repeated.

  Subsequently, a process of referring to the table T2 is performed in step S13. In step S13, the bit allocation Ba and the reference value R are read from the RAM 25, and a shift amount S2 and conversion coefficients C1 and C2 are generated with reference to the conversion table T2 stored in the ROM. The shift amount S2 and the conversion coefficients C1 and C2 are stored in the RAM 25. Since conversion table T2 is similar to conversion table T2 described in FIG. 2, description thereof will not be repeated.

Subsequently, a conversion process is performed in step S14. In the conversion process, the conversion coefficients C1 and C2 and the quantization code X are read from the RAM 25, the addition and multiplication processes shown in the following equation (6) are performed, and the conversion value Z is generated. The generated conversion value Z is stored in the RAM 25.
Z = C1 * (X + C2) (6) Re-display Subsequently, a shift process is performed in step S15. In the shift process, the shift amounts S1 and S2 and the conversion value Z are read from the RAM 25, and the operation shown in the following equation (22) is further performed to generate the inverse quantization value Y. . The generated inverse quantization value Y is stored in the RAM.
Z2 = Z * 2 ^(S1) (22) reprinted Y = Z2 * 2 ^(S2) (23) reprinted As described above, the inverse quantization and inverse scaling processes are performed, and this process ends in step S16. .

  As shown in the second embodiment, the present invention can be applied even when a computer is used to perform inverse quantization and inverse scaling processing by software.

[Examination of improvement of calculation accuracy]
Hereinafter, taking a specific numerical value as an example, a case will be described in which the inverse quantization and inverse scaling processing according to the present invention improves the calculation accuracy as compared with the conventional method.

  First, it is assumed that an operation is performed using a DSP equipped with a 24-bit fixed-point arithmetic architecture.

  The specifications of this DSP will be described. The access unit to the RAM and ROM is 24 bits long, and values such as table data stored in the ROM and calculation results stored in the RAM are treated as 24-bit data.

  The bit length of the register in the DSP is 24 bits, and the accumulator for storing the operation result is 56 bits. In addition and multiplication, the data in the two registers is input, and the operation result is output to the accumulator. When storing the result of the operation in the RAM or register to perform the next operation, the 24 bits at an arbitrary location in the 56 bits of the accumulator are extracted and stored. That is, the DSP can perform addition and multiplication such that a 24-bit value is input and a 24-bit value is output. In both addition and multiplication, the most significant bit in 24 bits is treated as a sign bit.

  By the way, when performing an operation using such a DSP, there are the following properties.

  First, since the value is handled as 24-bit data of a fixed decimal point, the range of values that can be expressed differs depending on the position of the decimal point to be set. For example, if the most significant bit is set as a sign bit, the subsequent 2 bits are set as an integer part, and the remaining 21 bits are set as a decimal part, the value range is -4 to +4.

  Second, since a value less than 2 to the power of −21 does not have a corresponding bit, it is approximated by rounding off, and a rounding error occurs. The rounding error becomes smaller as the decimal point position is set to the upper bit side, but overflow tends to occur. On the contrary, as the decimal point position is set to the lower bit side, overflow is prevented, but the rounding error becomes larger.

  Because of this property, in order to perform calculation with high accuracy, it is necessary to set the position of the decimal point so that the rounding error is as small as possible without causing overflow of digits.

  Assuming the DSP as described above, the inverse quantization and inverse scaling processing are performed by the conventional method as follows. In the conventional method, processing is performed according to the following equation.

Y = Sv * P2 [Ba] * (X + P3 [Ba]) (36)
The range of each value generated in this calculation process and the setting of the optimum decimal point have a great influence on the error.

  FIG. 6 is a diagram showing a range of values and a decimal point position generated in the calculation process of the calculation by the conventional method.

Referring to FIG. 6, the value of scaling factor Sv was obtained by calculating Sv = 2 ^{((3-Scf) / 3)} when scale factor index Scf was in the range of ^0-62 . Further, the ranges of P2 [Ba] and P3 [Ba] were obtained from FIG.

  FIG. 7 is a diagram showing a calculation result calculated to calculate the maximum value and the minimum value of FIG.

  In FIG. 7, each value of X, (X + P3 [Ba]), P2 [Ba] * (X + P3 [Ba]) is determined based on the X creation method described in the description of the background art and FIG. The maximum value and the minimum value for the numbers are actually calculated.

  FIG. 8 is a diagram comparing a theoretical value that does not consider the occurrence of an error and a calculated value when the conventional method is used.

  FIG. 8 shows an example of calculated values when the number of steps = 9, X = 0.0, and Scf = 1 (corresponding to Sv = 1.58740105196820). In the conventional method, the inverse quantization value Y causes an operation error of about 0.000000256 (= 1.4110231572988900−1.4110229015350300) with respect to the theoretical value. One reason for the large difference from the theoretical value of Sv is that the fractional bits of Sv in FIG.

In order to compare with this, an error due to the calculation used in the present invention is examined. In the present invention, processing is performed according to the following equation.
Y = (C1 * (X + P3 [Ba])) >> (S1 + S2) (37)
However, >> indicates a bit shift operator. In the above equation, each variable is
C1 = (2 ^{(mod (3-Scf, 3) / 3)} * P2 [Ba]) * (2 ^(-S2) ) (38)
S2 = int ((mod (3-Scf, 3) / 3) + log ₂ (P2 [Ba]) (39)
S1 = int ((3-Scf) / 3) (40)
It is.

  FIG. 9 is a table summarizing the range of each value generated in the calculation process of the present invention and the setting of the optimum decimal point position.

In FIG. 9, the reason why X, P3 [Ba], (X + P3 [Ba]), and Y are in the range shown in FIG. 9 is the same as in the case of the conventional calculation shown in FIG. . The reason for the range of C1 is that the shift amount for S2 to make (2 ^{(mod (3-Scf, 3) / 3)} * P2 [Ba]) less than 2 is indicated, and C1 is only S2. This is because it is defined as a shifted value.

  The reason why the range of (C1 * (X + P3 [Ba])) is not described in FIG. 9 is that the shift process of (S1 + S2) when obtaining Y can be realized by adjusting the position for extracting 24-bit data from the accumulator. It is. In other words, Y can be calculated while the value of (C1 * (X + P3 [Ba])) is held in the accumulator, so there is no need to set the decimal point position.

  FIG. 10 is a diagram comparing the calculated value calculated by the calculation method of the present invention with the theoretical value.

  FIG. 10 shows the result of the calculation corresponding to FIG. 8 with the number of steps = 9, X = 0.0, and Scf = 1.

  In FIG. 10, the inverse quantization value Y has an operation error of about 0.0000000017 (= 1.4110231572988900−1.4110231399536100) with respect to the theoretical value. The calculation error is suppressed to 1/10 or less of the calculation error as compared with the conventional method shown in FIG. Therefore, in the present invention, it is possible to reduce a calculation error when calculating the inverse quantization value Y.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

3 is a block diagram illustrating an exemplary configuration of an audio decoding device according to Embodiment 1. FIG. FIG. 2 is a block diagram illustrating a configuration of an inverse quantization / inverse scaling unit 3 in FIG. 1. FIG. 10 is a block diagram illustrating an exemplary configuration of an audio decoding device 21 according to a second embodiment. Fig. 4 is a flowchart for explaining a decoding process performed in the audio decoding device shown in Fig. 3. 5 is a flowchart showing in more detail the inverse quantization / inverse scaling process of step S4 in FIG. It is the figure which showed the range of the value produced in the calculation process of the calculation by the conventional method, and the position of the decimal point. It is a figure which shows the calculation result calculated in order to calculate the maximum value of FIG. 6, and the minimum value. It is the figure which compared the theoretical value which does not consider generation | occurrence | production of an error, and the calculated value at the time of performing by the conventional method. It is the figure which put together the range of each value which arises in the calculation process of this invention, and the setting of the optimal decimal point position. It is the figure which compared the calculated value calculated with the calculation method of this invention, and the theoretical value. Table B. of ISO / IEC 11172-3. It is the figure which showed an example of 2. It is the figure which showed the response | compatibility with bit allocation Ba and the number of steps (Number of steps). Table B. of ISO / IEC 11172-3. FIG. It is a schematic block diagram for demonstrating the conventional inverse quantization and an inverse scaling apparatus. It is a flowchart in the case of performing the conventional inverse quantization and inverse scaling process using computers, such as DSP.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Audio decoding apparatus, 2 Format analysis part, 3 Inverse quantization and inverse scaling part, 4 Filter part, 10, 11 Table reference part, 12 Conversion part, 13 Shift processing part, 21 Audio decoding apparatus, 22 Input I / O F part, 23 operation part, 26 output I / F part, 110,111 table reference part, 112 conversion part, 113 conversion part, T1, T2 table.

Claims (9)

An inverse quantization device that receives a quantization code X, a scale factor index Scf, and a bit allocation Ba, performs inverse quantization and inverse scaling processing, and outputs an inverse quantization value Y;
First storage means for storing a shift amount S1 and a reference value R corresponding to each scale factor index Scf calculated in advance based on the following equations (1) and (2);
Referring to the first storage means, first reference means for obtaining the shift amount S1 and the reference value R corresponding to the inputted scale factor index Scf;
Second storage means for storing a shift amount S2 corresponding to each reference value R and each bit allocation Ba calculated in advance based on the following equations (3), (4), and (5), and transform coefficients C1 and C2;
The second storage means is used to obtain the shift amount S2, the conversion coefficients C1 and C2 corresponding to the reference value R output from the first reference means and the input bit allocation Ba. Means for referencing,
Conversion means for outputting the converted value Z no line conversion according to the following equation (6) by using the transform coefficients C1 and C2 outputted input quantization code X from the second reference means,
Wherein said first to said converted value Z output from the conversion means, second of the shift amount obtained respectively by the reference means S1, S2 no line shift processing according to the following equation (7) using the inverse quantized value Inverse quantization apparatus comprising shift processing means for outputting Y:
S1 = int (P1 [Scf] / M) (1)
R = mod (P1 [Scf], M) (2)
S2 = int ((R / M) + log ₂ (P2 [Ba]) (3)
C1 = (2 ^{(R / M)} * P2 [Ba]) * (2 ^(-S2) ) (4)
C2 = P3 [Ba] (5)
Z = C1 * (X + C2) (6)
Y = Z * 2 ^{(S1 + S2)} (7)
However, P1 [Scf] indicates an integer predetermined for each scale factor index Scf, and P2 [Ba] and P3 [Ba] indicate a predetermined decimal number for each bit allocation Ba value. M represents a predetermined positive integer.   The shift processing means performs the shift processing by performing a first shift operation corresponding to the shift amount S1 and further performing a second shift operation corresponding to the shift amount S2. The inverse quantization apparatus according to 1.   2. The inverse quantization apparatus according to claim 1, wherein the shift processing unit performs the shift processing by collectively performing a shift operation corresponding to a total value of the shift amount S <b> 1 and the shift amount S <b> 2. The second storage means stores a conversion coefficient C2 calculated in advance based on the following equation (8) instead of the equation (5),
The said conversion means produces | generates the said conversion value Z according to the following Formula (9) instead of Formula (6) based on the conversion coefficients C1 and C2 and the quantization code | cord | chord X. Inverse quantization device.
C2 = C1 * P3 [Ba] (8)
Z = C1 * X + C2 (9) The P1 [Scf] is 3-Scf,
The inverse quantization apparatus according to claim 1, wherein M is three. An audio decoding device that performs a decoding process based on an audio encoding method including an inverse quantization process at the time of decoding,
An audio decoding device comprising the inverse quantization device according to claim 1. An image decoding apparatus that performs a decoding process based on an image encoding method including an inverse quantization process at the time of decoding,
An image decoding apparatus comprising the inverse quantization apparatus according to claim 1. A inverse quantization method of outputting the quantization code X, the scale factor index Scf and bit allocation are no rows inverse quantization and inverse scaling processes undergoing Ba dequantized value Y,
Referring to the first storage means for storing the shift amount S1 and the reference value R corresponding to each scale factor index Scf calculated in advance based on the following equations (1) and (2), the inputted scale factor index Scf Obtaining the shift amount S1 and the reference value R corresponding to
Refer to second storage means for storing shift values S2 and transform coefficients C1 and C2 corresponding to each reference value R and each bit allocation Ba calculated in advance based on the following equations (3), (4), and (5) And obtaining the shift amount S2, the conversion coefficients C1 and C2 corresponding to the obtained reference value R and the input bit allocation Ba,
Performing conversion according to the following equation (6) using the input quantization code X and the obtained conversion coefficients C1 and C2, and outputting a conversion value Z;
A step of performing a shift process according to the following equation (7) using the shift amounts S1 and S2 obtained for the output converted value Z and outputting the dequantized value Y:
S1 = int (P1 [Scf] / M) (1)
R = mod (P1 [Scf], M) (2)
S2 = int ((R / M) + log ₂ (P2 [Ba]) (3)
C1 = (2 ^{(R / M)} * P2 [Ba]) * (2 ^(-S2) ) (4)
C2 = P3 [Ba] (5)
Z = C1 * (X + C2) (6)
Y = Z * 2 ^{(S1 + S2)} (7)
However, P1 [Scf] indicates an integer predetermined for each scale factor index Scf, and P2 [Ba] and P3 [Ba] indicate a predetermined decimal number for each bit allocation Ba value. M represents a predetermined positive integer.   An inverse quantization program for causing a computer to execute processing based on the inverse quantization method according to claim 8.

Images