KR100202308B1 - Desibel converter and reverse converter - Google Patents

Abstract

The present invention relates to a decibel conversion method and an inverse conversion method of a decibel applied to a processor core of an MPEG-2 audio encoder. In particular, the present invention relates to a method for efficiently implementing a MPEG-2 audio encoder with a small number of tables. The present invention relates to a decibel and inverse transform method that can be calculated.

The present invention relates to a step of normalizing a power value having an FFT of 32 bits and associating an mantissa and an exponent, extracting a predetermined upper bit among the 32 bits of mantissa, and normalizing and calculating the extracted upper bit value. Obtaining a calculated exponent value from the mantissa table and the exponent table, adding a value obtained from the mantissa table and a value obtained from the exponent table, and an output value of a 16-bit 8.8 format. Multiplying a value of 21771, which is an approximation, to create a 32-bit value, extracting a predetermined bit from the 28th bit into an integer part i ₂ from the operation value calculated by the multiplier, and starting from the next bit of the integer part Extracting a predetermined bit into a fractional part, obtaining a value corresponding to the decimal part from a decimal table, and shifting a value obtained from the decimal part table by 31-i ₂ bits to the right It provides a decibel inverse transform method comprising the step.

Description

Decibel Conversion and Inverse Conversion Methods

The present invention relates to a decibel conversion method and an inverse conversion method of a decibel applied to a processor core of an MPEG-2 audio encoder. In particular, the present invention relates to a method for efficiently implementing a MPEG-2 audio encoder with a small number of tables. The present invention relates to a decibel and inverse transform method that can be calculated.

In the MPEG-2 audio coding process, many calculations are required to obtain a psychoacoustic model for bit allocation. In particular, the decibels ) And the inverse transformation of decibels ( Because we do a lot of computation, we need a fast decibel operator and an inverted decibel operator. However, in general, transcendental functions such as logs or exponents require a large number of cycles to calculate once, using a digital signal processor (DSP), for example, in calculations requiring high precision. The transcendental function is calculated after about 10 clocks.

However, if the psychoacoustic model is implemented under such conditions, real-time implementation of the MPEG-2 audio encoder is impossible. Therefore, it is important to quickly calculate the psychoacoustic model in the real-time implementation of the MPEG-2 audio encoder, and the time used for calculating the log and the exponent is an important factor for the real-time implementation of the MPEG-2 audio encoder.

However, the process of obtaining a psychoacoustic model does not require a very high precision because it does not actually produce a transmitted signal, and the transcendental function can obtain a value using a table, so that the result can be obtained in fewer cycles. .

1 is a block diagram for the implementation of the decibel calculation method of the present invention.

Figure 2 is a block diagram for the implementation of the decibel inverse transformer method of the present invention.

Decibel calculation using a table in MPEG outputs a 16-bit output by inputting a 32-bit signed fast fourier transform (FFT) signal power (= x) value. Create The input x is a 32 bit unsigned number and the output y takes a signed number in 16 bit 8.8 format. The general operation procedure using such a table is as follows.

The mantissa and exponent are found by normalizing the 32-bit input x, the FFT power value. In this case, the mantissa is a value shifted x to the left so that the most significant bit is 1, and the exponent is the number of shifts. In other words,

ego,

The decibel value of x, y

to be. Therefore, from the mantissa table From the index table Find and sum y to find y. In this case the same procedure get a decibel is a singer (mantissa) is up to 32 bits, the most significant bit is because it is always 1 and requires 2 to ³¹ table, index (exponent) is required to 32 table owing to the 32 value Do. In this case, the highest precision can be obtained, but it is difficult to implement in hardware because the number of tables is too large. In addition, the process of obtaining a psychoacoustic model does not require such high precision since it does not actually produce a transmitted signal.

Thus, in these two aspects, it is necessary to calculate the decibels with fewer tables. In view of the above, the present invention implements the decibel operator used in the MPEG-2 audio coder as shown in FIG. 1 with 32 (2 ⁵ ) mantissa table values and 32 index table values. The mantissa table is made with the top 5 bits except the most significant bit of the mantissa, and the exponent table is made with the exponent. The process of creating such a table is as follows.

The decibel value of power x, y

Here, mantissa_table [mantissa] represents a table value corresponding to mantissa, and exponent_table [exponent] represents a table value corresponding to exponent.

In other words, the mantissa table value is

mantissa_table [mantissa] 10 log ₁₀ (mantissa 2 ^-26 ) 2 ⁸ ,

exponent table value is

exponent_table [exponent] (26-exponent) 10 log ₁₀ 2 2 ⁸

To create.

mantissa 2 ^-26 is mantissa [31: 0] 2 ^-26 , which can be expressed as mantissa [31:26]. Here, the expression of [a: b] means the most significant a bit to the least significant b bit. And if 5 bits except double most significant bit is manti,

mantissa [31:26] = 2 ⁵ + mantissa [30:26]

= 2 ⁵ + manti

Because of,

mantissa_table [manti] = 10 log ₁₀ (2 ⁵ + manti) 2 ⁸

exponent_table [exponent] = (26-exponent) 10 log ₁₀ 2 2 ⁸

Can be expressed as Table 1 and Table 2 produced by such a process is as follows.

The decibel calculator shown in FIG. 1 performs a decibel operation by the following procedure. First, mant is calculated by taking the mantissa and the exponent from the 32-bit input x and taking the upper five bits except the most significant bit of the mantissa.

Next, if x is 0, print y = -25600, otherwise

Find y by adding mantissa_table [manti] and exponent_table [exponent].

On the other hand, the inverse conversion of the decibel used in the MPEG-2 audio coder, like the decibel conversion, does not affect the entire masking curve even if there is some error, and can be calculated using a simple table.

Inverse transformation uses 32-bit power by inputting the decibel value (y) in 16-bit 8.8 format Is the process of creating. The operation of the inverse transform is as follows.

If y ₁ is defined as

Power x is

It is expressed as

Assuming y ₁ = i + f, where

i: integer part of y ₁

f: fractional part of y ₁

to be. therefore,

to be. This value is shifted 2 ^f left by i times. At this time, 2 ^f uses a table. The process of implementing integer operation using this process is as follows.

Define y ₂ as

We define i ₂ and f ₂ as

That is, the actual data value

Because of

to be.

So x is

Because of, Shift to the right by 31-i ₂ bits. The table used at this time is

Use 2 is a decibel inverse transformer made by implementing such a method and performs the operation as follows.

First, 16 bits of y are multiplied by 21771 to output 32 bits. From the output

After applying the value of, find fract_table [f ₂ ] and shift it to the right by 31-i ₂ bits.

The present invention enables real-time implementation of an MPEG-2 audio encoder by performing decibel operations and inverse transform operations in decibels from a small number of mantissa and exponent tables as described above.

In addition, a preferred embodiment of the present invention is disclosed for the purpose of illustration, those skilled in the art will be able to make various modifications, changes, additions, etc. within the spirit and scope of the present invention, such modifications and modifications belong to the following claims You will have to look.

Claims (4)

Normalizing the power value of the 32-bit FFT to calculate a mantissa and an exponent; extracting a predetermined upper bit among the 32-bit mantissa; and an exponent value calculated by normalizing with the extracted upper order bit value. And obtaining from the mantissa table and the exponent table, and adding the value obtained from the mantissa table and the value obtained from the exponent table. The decibel conversion method according to claim 1, wherein the predetermined high order bits are high order five bits excluding the most significant bit. 16-bit 8.8 format output Multiplying the value 21771, which is an approximation of 2 ¹⁶ , to form a 32-bit value, extracting a predetermined bit from the 28th bit into an integer part i ₂ from the operation value calculated by the multiplier, and the next bit of the integer part Extracting the predetermined bit into a fractional part, obtaining a value corresponding to the decimal part from a decimal table, and calculating a value obtained from the decimal part table by 31-i ₂ bits to the right. The decibel inverse transform method comprising the step of shifting. 4. The method of claim 3, wherein the predetermined bit is five bits.