JP2005197989A - Arithmetic circuit for exponentiation, quantization circuit and its method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the processing speed of quantization processing and to improve the quality of a reproduced signal. <P>SOLUTION: When an output signal mc from an MDCT circuit is a threshold K and more, "exponentiation approximate value" (pow_spec_av) is calculated by using a linear equation based on respective coefficients and shift values which are respectively acquired from a 1st coefficient storage part 603, a 1st shift value storage part 605 and a 2nd coefficient storage part 607, and a 2nd shift value storage part 609. When the signal mc is less than the threshold K, a "exponentiation value" (pow_spec_ac) stored in a exponentiation value storage part 611 is acquired. A exponentiation value corresponding to the signal mc less than the threshold K is previously calculated and stored in the exponentiation value storage part 611. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a quantization circuit in a signal compression apparatus.

  MPEG (Moving Picture Experts Group) is used as a method for compressing image signals and audio signals. MPEG converts a time-domain input signal into a frequency-domain signal, and generates a compressed signal by quantizing and encoding coefficients in each frequency band. Also, MPEG4 / AAC (Advanced Audio Coding) is used as an audio signal compression algorithm.

JP-A-8-101698

  When decoding and reproducing a compressed image signal or audio signal, the quality of the image or audio is affected by the accuracy of quantization. However, in order to increase the accuracy of quantization, the processing load is increased and the circuit scale is also increased. In addition, the time required for processing becomes longer. In particular, the non-linear quantization process needs to perform a power operation and further increases the processing load.

  In the above-mentioned Patent Document 1, a technique for performing simple quantization using a polygonal line table is disclosed, but since all signals are quantized using the polygonal line table, there remains a problem from the viewpoint of quantization accuracy. Yes.

  In view of the above problems, an object of the present invention is to reduce the quantization processing time and improve the quality of a reproduction signal.

  In order to solve the above-mentioned problem, the invention according to claim 1 is characterized in that: input means for inputting a coefficient signal for each frequency band of the frequency-divided input signal; means for comparing the coefficient signal with a predetermined threshold; A power value storage means for storing a power value of the coefficient signal in advance, and a means for acquiring a power value of the coefficient signal with reference to the power value storage means when the coefficient signal falls below a predetermined threshold value And approximating means for performing approximation calculation using a linear equation to calculate a power approximation value of the coefficient signal when the coefficient signal exceeds a predetermined threshold value.

  According to a second aspect of the present invention, in the power operation circuit according to the first aspect, the power value storage means stores only a power value corresponding to a coefficient signal below the threshold value.

  According to a third aspect of the present invention, in the power arithmetic circuit according to the first or second aspect, the coefficient signal is further divided into a plurality of sections, and equations for storing linear equation information corresponding to the respective sections are stored. Information storage means, wherein the approximation means obtains a linear equation corresponding to the input coefficient signal by referring to the equation information storage means, and calculates the power approximation value.

  According to a fourth aspect of the present invention, the step of inputting a coefficient signal for each frequency band of the frequency-divided input signal, and when the coefficient signal falls below a predetermined threshold, a power value of the coefficient signal is stored in advance. The power value of the coefficient signal is obtained by referring to the power value storage means, and when the coefficient signal exceeds a predetermined threshold, approximation calculation using a linear equation is performed to perform power approximation of the coefficient signal And a step of calculating a value.

  According to a fifth aspect of the present invention, there is provided input means for inputting a coefficient signal for each frequency band of the frequency-divided input signal, means for comparing the coefficient signal with a predetermined first threshold, and a power value of the coefficient signal And a power value of the coefficient signal (hereinafter, referred to as the first power value storage means) with reference to the first power value storage means when the coefficient signal falls below a predetermined first threshold value. When the coefficient signal exceeds a predetermined first threshold value, an approximation calculation using a linear equation is performed to calculate a power approximation value of the coefficient signal (hereinafter referred to as a first power value). An approximation means for calculating the first power value, a means for obtaining a quantized value based on the first power value or the first power approximate value, the quantized value and a predetermined second threshold value. Means for comparing and a power of the quantized value Is stored in advance, and when the quantized value falls below a predetermined second threshold value, the second power value storing means is referred to and the quantized power value ( Hereinafter, when the quantized value exceeds a predetermined second threshold value, an approximation calculation using a linear equation is performed to obtain a power approximate value of the quantized value. An approximation means for calculating (hereinafter referred to as a second power approximation value) and a means for obtaining a dequantization value based on the second power value or the second power approximation value.

  According to a sixth aspect of the present invention, there is provided a step of inputting a coefficient signal for each frequency band of the input signal divided into frequency bands, and a power value of the coefficient signal when the coefficient signal falls below a predetermined first threshold value. Is stored in advance, the power value of the coefficient signal (hereinafter referred to as the first power value) is acquired, and the coefficient signal exceeds a predetermined first threshold value. Includes calculating a power approximation value of the coefficient signal (hereinafter referred to as a first power approximation value) by performing an approximation calculation using a linear equation, and the first power value or the first power approximation value. A step of obtaining a quantized value on the basis of the second power value storage means storing in advance a power value of the quantized value when the quantized value is below a predetermined second threshold value , The power value of the quantized value (hereinafter referred to as the second When the quantized value exceeds a predetermined second threshold value, an approximate calculation using a linear equation is performed to calculate a power approximate value of the quantized value (hereinafter referred to as a second power). And calculating the inverse power value based on the second power value or the second power approximation value.

  According to the present invention, the “power value” calculated in advance is used in a range where high accuracy is required, and the “power value” is calculated by approximate calculation using a linear equation in the other ranges. It is possible to improve the processing speed of the quantization process and improve the quality of the reproduction signal.

{1. Overall process flow}
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a schematic diagram of an apparatus that performs audio data compression processing based on the MPEG4 / AAC algorithm. This apparatus inputs an MDCT (Modefied Descrete Cosine Transform) circuit 1 that converts an input signal (audio signal) in the time domain into a signal in the frequency domain, and a signal output from the MDCT circuit 1 and quantizes the signal. A quantization / encoding circuit 2 that repeatedly performs quantization and encoding processing until the noise is less than an allowable amount and the code amount is less than or equal to a predetermined bit rate. The quantization / encoding circuit 2 includes two circuit blocks, a quantization circuit 3 and an encoding circuit 4.

  FIG. 2 is a diagram illustrating a block configuration of the quantization circuit 3. The quantization circuit 3 includes a quantization operation circuit 5, an inverse quantization operation circuit 10, and a noise determination circuit 11. Furthermore, the quantization arithmetic circuit 5 includes a quantization power arithmetic circuit 6 and a quantization coefficient arithmetic circuit 7, and the inverse quantization arithmetic circuit 10 includes an inverse quantization power arithmetic circuit 8 and an inverse quantization coefficient arithmetic circuit 9. ing.

  In such a configuration, the signal (mc) output from the MDCT circuit 1 is input to the quantization arithmetic circuit 5. Specifically, the signal (mc) is a coefficient (MDCT coefficient) in each frequency band of the input signal (audio signal) frequency-divided by the MDCT circuit 1. In the quantization operation circuit 5, first, the signal (mc) is input to the quantization power operation circuit 6, where “power operation” is performed. The power calculation is expressed by, for example, Equation 1 where x is the input signal (mc) and y is the signal after the power calculation.

  When the “power operation value” of the signal (mc) is obtained in the quantization power operation circuit 6, the power operation value is then divided by the scale factor in the quantization coefficient operation circuit 7 and the quantized value (quant). Is calculated. As described above, the quantization operation circuit 5 performs nonlinear quantization processing.

  The quantized value (quant) is then input to the inverse quantization arithmetic circuit 10. In the inverse quantization operation circuit 10, first, in the inverse quantization power operation circuit 8, the “power value” of the quantized value (quant) is calculated. The power calculation is expressed by, for example, Equation 2 where q is the input signal (quant) and y ′ is the signal after the power calculation.

  When the “quantized power value” of the quantized value is obtained in the inverse quantization power calculation circuit 8, next, the inverse quantization coefficient calculation circuit 9 divides the power calculation value by the scale factor to obtain the inverse quantization value. (Requant) is calculated. As described above, the nonlinear quantization processing is also performed in the inverse quantization arithmetic circuit 10.

  The inverse quantized value (requant) is input to the noise determination circuit 11 to determine whether or not the quantization noise is within an allowable range. If the allowable noise amount is exceeded, the value of the scale factor is changed, and the quantized value and the inverse quantized value are calculated again. Then, noise determination is performed again using the newly obtained inverse quantization value (requant). Such a process is repeated until the quantization noise falls within the allowable range.

  When the quantization noise falls within the allowable range, Huffman coding is performed in the coding circuit 4 using the quantized value (quant) output from the quantization circuit 3. As a result of encoding, when the code amount does not satisfy the target bit rate, the scale factor is changed again, and a noise determination processing loop is performed in the quantization circuit 3. Such a process is repeated, and when the encoding amount reaches the target bit rate, the quantization / encoding process loop ends.

{2. Quantization processing}
Next, the configuration of the quantization arithmetic circuit 5 and the quantization arithmetic processing will be described with reference to FIGS.

  The quantization power calculation circuit 6 receives the signal (mc) output from the MDCT circuit 1. The signal (mc) is branched in seven directions, and the comparison circuit 601, the first coefficient acquisition unit 602, the first shift value acquisition unit 604, the second coefficient acquisition unit 606, the second shift value acquisition unit 608, and the power conversion unit. 610 and multiplier 612.

  The first coefficient acquisition unit 602, the first shift value acquisition unit 604, the second coefficient acquisition unit 606, and the second shift value acquisition unit 608 are based on the input signal (mc) range, respectively. Coefficients or shift values are acquired from 603, the first shift value storage unit 605, the second coefficient storage unit 607, and the second shift value storage unit 609. In the present embodiment, the signal (mc) output from the MDCT circuit 1, that is, the MDCT coefficient is an integer from 1 to 67108863 in decimal.

  FIG. 4 shows each coefficient and shift value stored in the first coefficient storage unit 603, the first shift value storage unit 605, the second coefficient storage unit 607, and the second shift value storage unit 609 in one table. It is. Therefore, in FIG. 3, the first coefficient storage unit 603, the first shift value storage unit 605, the second coefficient storage unit 607, and the second shift value storage unit 609 are shown as different hardware configurations. These storage units may be stored as a single table as shown in FIG. 4 using a common ROM.

  Although omitted in the table of FIG. 4, the index is associated with the value of the signal (mc). As shown in the figure, the value of the signal (mc) is divided into 89, and each section is indexed. The first coefficient, the first shift value, the second coefficient, and the second shift value are stored corresponding to each index. These coefficients and shift values are information defining a linear equation for approximate calculation. The range of the signal (mc) input to the quantization power arithmetic circuit 6 is an integer from 0 to 67108863 in decimal as described above. When the value of the signal (mc) is less than K, the power conversion unit 610 performs “power conversion”. When the value of the signal (mc) is greater than or equal to K, that is, when the value of the signal (mc) is an integer from K to 67108863, a linear equation corresponding to the index to which the signal (mc) belongs is determined. Then, an approximate calculation using the linear equation is performed to obtain a “power approximation value”.

  Each coefficient or each shift value shown in FIG. 4 is an example, and an optimal value may be set as appropriate by simulation or the like. Further, here, the signal (mc) is divided into 89 indexes, but the number of divisions of this index may be set as appropriate by simulation or the like. Further, the number of signals included in each index need not be constant. What is necessary is just to set so that a quantization error may become the minimum. Also, the numerical value represented as “first shift value” in FIG. 4 is a value obtained by adding 1 to the numerical value stored in the first shift value storage unit 605. This 1 addition process is a process in the adder 613 described later. An integer of up to 16 may be stored in the first shift value storage unit 605 as the first shift value, but here, in order to reduce the capacity of the first shift value storage unit 605 (to store it in 4 bits). ) Is stored with an integer of up to 15, and 1 is added in the subsequent processing. However, in FIG. 4, in order to express the original shift value, the numerical value after 1 is shown for convenience.

  When the value of the signal (mc) is equal to or greater than K, the first coefficient acquisition unit 602, the first shift value acquisition unit 604, the second coefficient acquisition unit 606, and the second shift value acquisition unit 608 The coefficient value or shift value is acquired from the first coefficient storage unit 603, the first shift value storage unit 605, the second coefficient storage unit 607, and the second shift value storage unit 609 according to the index to which the For example, when the index is 10, the first coefficient acquisition unit 602 sets “825” as the first coefficient, and the first shift value acquisition unit 604 sets “16” as the first shift value (as described above, actually 1 is added after acquiring “15”), the second coefficient acquisition unit 606 sets “1652” as the second coefficient, and the second shift value acquisition unit 608 sets “5” as the second shift value. To get.

  On the other hand, when the value of the signal (mc) is less than K, the power conversion unit 610 stores the “power value” stored in a one-to-one correspondence with the value of the signal (mc) from the power value storage unit 611. And “power value” (pow_spec_ac) is output to the selector 617. The power value storage unit 611 stores a value obtained by calculating in advance a “power value” for a signal (mc) of less than K. Therefore, the power value storage unit 611 only needs to have a capacity capable of storing K “power values”.

  Next, the multiplier 612 receives the signal (mc) and the first coefficient (q_a_int) output from the first coefficient acquisition unit 602, multiplies these two signals, and outputs a multiplication result (tmp_1). The adder 613 receives the first shift value (q_a_sft) output from the first shift value acquisition unit 604 and adds 1 to the first shift value. This 1-added shift value is used as an added shift value (q_a_sft + 1) (that is, the “first shift value” shown in FIG. 4 is actually an added shift value (q_a_sft + 1). .) Then, the shifter 614 receives the multiplication result (tmp_1) and the addition shift value (q_a_sft + 1), and performs a right shift operation on the multiplication result (tmp_1) by the addition shift value (q_a_sft + 1). Specifically, when the multiplication result (tmp_1) is expressed by 2 bits, the bit shift is performed by the decimal integer indicated by the addition shift value (q_a_sft + 1). For example, in the case of the index 10, when the multiplication result (825 × (mc)) is expressed by 2 bits, a 16 (15 + 1) bit right shift operation is performed. The shifter 614 outputs the signal (tmp_2) after the right shift calculation to the adder 616.

  Further, the shifter 615 receives the second coefficient (q_b_int) output from the second coefficient acquisition unit 606 and the second shift value (q_b_sft) output from the second shift value acquisition unit 608, and inputs the second coefficient (q_b_int). Left shift operation is performed by the second shift value (q_b_sft). Specifically, when the second coefficient (q_b_int) is expressed by 2 bits, the bit shift is performed by the decimal integer indicated by the second shift value (q_b_sft). For example, in the case of the index 10, when the second coefficient (1652) is expressed by 2 bits, a 5-bit left shift operation is performed. The shifter 615 outputs the signal (tmp_3) after the left shift calculation to the adder 616.

  Next, the adder 616 receives the output signal (tmp_2) from the shifter 614 and the output signal (tmp_3) from the shifter 615, and adds the two signals. Then, the adder 616 outputs a “power approximation value” (pow_spec_av) to the selector 617.

  The selector 617 receives the “power approximation value” (pow_spec_av) output from the adder 616 and the “power value” (pow_spec_ac) output from the power conversion unit 610.

  The comparison unit 601 receives the signal (mc) and compares the signal (mc) with the threshold value K. When the value of the signal (mc) is less than the threshold value K, a control signal “1” is transmitted to the selector 617 so as to output a “power value” (pow_spec_ac). When the value of the signal (mc) is equal to or greater than the threshold value K, the “power approximation value” (pow_spec_av) that is the result of the approximation calculation using the first coefficient, the first shift value, the second coefficient, and the second shift value. ) Is transmitted to the selector 617 so as to output. The selector 617 outputs either the “power approximation value” (pow_spec_av) or the “power value” (pow_spec_ac) as the “power operation value” (pow_spec) based on the control signal input from the comparison unit 601.

  The “power calculation value” (pow_spec) output from the quantization power calculation circuit 6 is input to the quantization coefficient calculation circuit 7. In the quantization coefficient calculation circuit 7, the “power calculation value” (pow_spec) is divided by a scale factor given from another circuit (not shown) to calculate a quantization value (quant). The calculated quantized value (quant) is output to the encoding circuit 4 and also output to the inverse quantization operation circuit 10. The range of the quantized value (quant) corresponding to the range 0 to 67108863 (decimal number) of the input signal (mc) is 0 to 8191 (decimal number).

  As described above, in the present invention, when the value of the input signal (mc) is less than the threshold value K, a highly accurate “power value conversion” is performed using a table in which the “power value” is calculated in advance, When the value of the signal (mc) is larger than the threshold value K, the power value is approximated by a method using a simple linear equation, so that the storage capacity of the storage unit composed of a ROM or the like as a whole is obtained. It can be made smaller.

  Here, the threshold value K may be obtained experimentally from the quality of the reproduced sound of the finally encoded audio data. However, in the range where the value of the signal (mc), that is, the value of the MDCT coefficient is small and the S / N ratio is large, the “power operation” with high accuracy should be performed. It is preferable to use a table and adopt the output of the power conversion unit 610. That is, as shown in FIG. 5, according to the present embodiment, in a range where the value of the signal (mc) is small and the S / N ratio is large, high-precision quantization is performed and the S / N ratio is small. In the range, simple approximation using a linear equation (y = ax + b) is performed, so it is possible to reduce the quantization processing load while maintaining high quality of the playback signal. It is.

{3. Inverse quantization process}
Next, the inverse quantization operation circuit 10 and the inverse quantization operation processing will be described with reference to FIGS. 6 and 7.

  The inverse quantization power operation circuit 8 receives the quantization value (quant) output from the quantization operation circuit 5. The quantized value (quant) is branched in five directions and input to the comparison circuit 801, the third coefficient acquisition unit 802, the fourth coefficient acquisition unit 804, the power conversion unit 806, and the multiplier 808.

  The third coefficient acquisition unit 802 and the fourth coefficient acquisition unit 804 acquire coefficient values from the third coefficient storage unit 803 and the fourth coefficient storage unit 805, respectively, based on the input quantized value (quant) range. To do.

  FIG. 7 shows each coefficient value stored in the third coefficient storage unit 803 and the fourth coefficient storage unit 805 in one table. Therefore, in FIG. 6, the third coefficient storage unit 803 and the fourth coefficient storage unit 805 are shown as different hardware configurations, but these storage units are shown in FIG. 7 using a common ROM. You may make it store as such a single table.

  Although omitted in the table of FIG. 7, the index is associated with the value of the quantized value (quant). As shown in the figure, the value of the quantized value (quant) is divided into 21 and each section is indexed. The third coefficient and the fourth coefficient are stored corresponding to each index. These coefficients are information defining a linear equation for approximate calculation. The range of the quantization value (quant) input by the inverse quantization power operation circuit 8 is an integer of 0 to 8191 as described above. When the quantized value (quant) is less than L, the power conversion unit 806 performs “power conversion”. When the value of the quantized value (quant) is L or more, that is, when the value of the quantized value (quant) is an integer of L to 8191, it corresponds to the index to which the quantized value (quant) belongs. The first-order equation is determined, and an approximate operation using the first-order equation is performed to obtain a “power approximation value”.

  Each coefficient shown in FIG. 7 is an example, and an optimal value may be set as appropriate by simulation or the like. Also, here, the quantized value (quant) is divided into 21 indexes, but the number of divisions of this index may be set as appropriate by simulation or the like. Further, the number of signals included in each index need not be constant. What is necessary is just to set so that a quantization error may become the minimum.

  When the value of the quantized value (quant) is greater than or equal to L, the third coefficient acquisition unit 802 and the fourth coefficient acquisition unit 804 respectively include a third coefficient storage unit according to the index to which the quantized value (quant) belongs. The coefficient value is acquired from 803 and the fourth coefficient storage unit 805. For example, when the index is 10, the third coefficient acquisition unit 802 acquires “12.75” as the third coefficient, and the fourth coefficient acquisition unit 806 acquires “2751” as the fourth coefficient.

  On the other hand, when the quantized value (quant) is less than L, the power conversion unit 806 stores the “power value” stored in the power value storage unit 807 in a one-to-one correspondence with the quantized value (quant). To get. The power value storage unit 807 stores a value obtained by calculating in advance a “power value” for a quantized value (quant) less than L. Therefore, the power value storage unit 807 only needs to have a capacity capable of storing L “power values”.

  Next, the multiplier 808 receives the quantized value (quant) and the first coefficient (iq_a) output from the first coefficient acquisition unit 802, multiplies these two signals, and outputs the multiplication result (tmp_4). To do. The shifter 809 receives the multiplication result (tmp_4) and performs a 2-bit right shift operation on the multiplication result (tmp_4). Specifically, when the multiplication result (tmp_4) is expressed by 2 bits, bit shift is performed by 2 bits. For example, in the case of the index 10, when the multiplication result (12.75 × (quant)) is expressed by 2 bits, the right shift operation is performed by 2 bits. The shifter 809 outputs the signal (tmp_5) after the shift operation to the subtracter 810.

  Next, the subtractor 810 inputs the signal (tmp_5) after the shift operation and the second coefficient (iq_b), subtracts the second coefficient (iq_b) from the signal (tmp_5) after the shift operation, Value "(pow_quant_av) is output. Then, the selector 811 receives “power calculation approximate value” (pow_quant_av) that is an output signal of the subtractor 810 and “power value” (pow_quant_ac) that is an output signal of the power conversion unit 610.

  The comparison unit 801 receives the quantized value (quant) and compares the quantized value (quant) with the threshold value L. When the value of the quantized value (quant) is less than the threshold L, the control signal “1” is transmitted to the selector 811 so as to output the “power value” (pow_quant_ac) that is the output signal of the power conversion unit 806. To do. When the value of the quantized value (quant) is equal to or greater than the threshold value L, the selector 811 is controlled to output the “power operation approximate value” (pow_quant_av) obtained by the approximate calculation using the third coefficient and the fourth coefficient. Signal “0” is transmitted. Based on the control signal input from the comparison unit 801, the selector 811 outputs either the “power calculation approximate value” (pow_quant_av) or the “power value” (pow_quant_ac) as the “power calculation value” (pow_spec). .

  The “power calculation value” (pow_spec) output from the inverse quantization power calculation circuit 8 is input to the inverse quantization coefficient calculation circuit 9. In the inverse quantization coefficient calculation circuit 9, the “power calculation value” (pow_spec) is divided by a scale factor given from another circuit (not shown) to calculate an inverse quantization value (requant). The calculated inverse quantization value (requant) is input to the noise determination circuit 11. In the noise determination circuit 11, quantization noise is calculated from the signal (mc) that is the MDCT coefficient and the inverse quantization value (requant), and whether or not the quantization noise satisfies a predetermined threshold, that is, the quantization noise is calculated. It is determined whether or not the amount is within the allowable range. This threshold is appropriately set according to the required quantization accuracy.

  When the quantization noise does not satisfy the predetermined threshold, that is, when the noise tolerance is exceeded, the scale factor is changed, and as described above, the quantization and inverse quantization processes are performed again. A determination of quantization noise is made. When the quantization noise satisfies a predetermined threshold, the encoding process is executed in the encoding circuit 4 using the quantized value (quant) output from the quantization operation circuit 5. Thereafter, as described above, the encoding circuit 4 performs the encoding process.

  As described above, even in the inverse quantization process, when the value of the input signal (quant) is less than the threshold value L, “power value conversion” with high accuracy using a table in which the “power value” is calculated in advance is used. When the value of the signal (quant) is larger than the threshold value L, the “power value” is obtained by an approximate calculation by a method using a simple linear equation. It is possible to reduce the storage capacity of the part.

  In the above embodiment, the present invention is applied to a quantization circuit in the MPEG4 / AAC algorithm. However, the apparatus and method of the present invention can be applied to a general quantization circuit and a quantization method. is there.

1 is an overall schematic diagram of a speech encoding apparatus according to an embodiment. It is a block diagram of a quantization circuit. It is a block diagram of a quantization power operation circuit. It is a figure which shows the power calculation approximation table. It is a figure which shows the approximation method of a power calculation. It is a block diagram of an inverse power operation circuit. It is a figure which shows an inverse power calculation approximation table.

Explanation of symbols

2 Quantization / Encoding Circuit 3 Quantization Circuit 4 Encoding Circuit 6 Quantization Power Operation Circuit 8 Inverse Quantization Power Operation Circuit

Claims (6)

Input means for inputting a coefficient signal for each frequency band of the frequency-divided input signal;
Means for comparing the coefficient signal with a predetermined threshold;
Power value storage means for storing a power value of the coefficient signal in advance;
Means for obtaining a power value of the coefficient signal with reference to the power value storage means when the coefficient signal is below a predetermined threshold;
When the coefficient signal exceeds a predetermined threshold value, an approximation unit that performs an approximate calculation using a linear equation to calculate a power approximation value of the coefficient signal;
A power operation circuit comprising: In the power operation circuit according to claim 1,
The power value storage means includes
Only a power value corresponding to a coefficient signal below the threshold value is stored. The power operation circuit according to claim 1 or 2, further comprising:
Equation information storage means for dividing the coefficient signal into a plurality of sections and storing information of a linear equation corresponding to each section;
With
The approximation means obtains a linear equation corresponding to the input coefficient signal by referring to the equation information storage means, and calculates the exponentiation approximation value. Inputting a coefficient signal for each frequency band of the frequency-divided input signal;
When the coefficient signal falls below a predetermined threshold value, the power value of the coefficient signal is obtained by referring to a power value storage means that stores a power value of the coefficient signal in advance. If the threshold is exceeded, performing an approximate calculation using a linear equation to calculate a power approximation of the coefficient signal; and
A power calculation method comprising: Input means for inputting a coefficient signal for each frequency band of the frequency-divided input signal;
Means for comparing the coefficient signal with a predetermined first threshold;
First power value storage means for storing a power value of the coefficient signal in advance;
Means for obtaining a power value of the coefficient signal (hereinafter referred to as a first power value) with reference to the first power value storage means when the coefficient signal falls below a predetermined first threshold;
Approximating means for performing approximation calculation using a linear equation and calculating a power approximation value of the coefficient signal (hereinafter referred to as a first power approximation value) when the coefficient signal exceeds a predetermined first threshold value; ,
Means for obtaining a quantized value based on the first power value or the first power approximation value;
Means for comparing the quantized value with a predetermined second threshold;
Second power value storage means for storing a power value of the quantized value in advance;
Means for obtaining a power value of the quantized value (hereinafter referred to as second power value) with reference to the second power value storage means when the quantized value is below a predetermined second threshold. When,
When the quantized value exceeds a predetermined second threshold, an approximation calculation using a linear equation is performed to calculate a power approximate value of the quantized value (hereinafter referred to as a second power approximate value). Means,
Means for obtaining a dequantization value based on the second power value or the second power approximation value;
A quantization circuit comprising: Inputting a coefficient signal for each frequency band of the input signal divided into frequency bands;
When the coefficient signal falls below a predetermined first threshold, the power value of the coefficient signal (hereinafter referred to as the first power) is referred to the first power value storage means that stores the power value of the coefficient signal in advance. When the coefficient signal exceeds a predetermined first threshold value, an approximate calculation using a linear equation is performed to calculate a power approximation value of the coefficient signal (hereinafter referred to as a first power approximation value). Calculating)
Obtaining a quantized value based on the first power value or the first power approximation value;
When the quantized value falls below a predetermined second threshold, the power value of the quantized value (hereinafter, referred to as a power value of the quantized value) When the quantized value exceeds a predetermined second threshold value, an approximate calculation using a linear equation is performed to calculate a power approximate value of the quantized value (hereinafter referred to as a second power value). Calculating a power-of-two approximation value),
Obtaining a dequantization value based on the second power value or the second power approximation value;
A quantization method comprising:

Images