JP2007241128A - Waveform data conversion apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To attain a waveform data conversion apparatus in which sound quality degradation caused by truncation is avoided, while keeping required resolution. <P>SOLUTION: In the data format conversion apparatus 22, a 24-bit waveform data of a fixed point form are converted to the waveform data of a floating point form, which is composed of a 3-bit exponent part to which discontiguous exponential is allocated, and a 13-bit significant part, so that a processing word length of a signal processing device 20 may be efficiently stored in accordance with the word length of an external memory 30, while this 16-bit waveform data of the floating point form are converted to 24-bit waveform data of the fixed point form. Thereby, the sound quality degradation by truncation of the waveform data is avoided, while keeping the required resolution. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a waveform data conversion apparatus that converts waveform data in a fixed-point format into a floating-point format.

  In order to reduce the amount of data, a technique for converting a data format from a fixed-point format to a floating-point format is known. As this type of technology, for example, Patent Document 1 discloses a technology in which waveform data is divided into exponent part data and mantissa part data to reduce the number of data bits (word length) per sample point.

JP-A-7-160267

  Incidentally, a signal processing device called a digital signal processor (DSP) is provided in a sound source of an electronic musical instrument. In signal processing apparatuses, waveform data is often delayed using an external memory. As in the technique disclosed in Patent Document 1, when the waveform data in the fixed-point format is converted to the floating-point format to reduce the data amount, the word length (storage width) of the external memory, the signal processing device It is necessary to harmonize the processing word length of the fixed-point format processed internally. This point will be described with reference to FIG.

  FIG. 7 is a diagram showing the conversion relationship between fixed-point 20-bit data and floating-point 16-bit data consisting of an exponent part 3 bits and a mantissa part 13 bits. In general, assuming that the word length of the exponent part in floating-point data is En and the word length (number of significant digits) of the mantissa part is Kn, the word length Xn of the fixed-point data can be expressed by an expression of Kn + 2 ^ En-1. Note that “^” in the above expression represents a power (power).

  Therefore, in the example shown in FIG. 7, the word length Xn of the fixed-point data corresponding to the change in the word length of the exponent part and the mantissa part of the floating-point 16-bit data is expressed by the equation 16−En + 2 ^ En−1. it can. Specifically, when En = 0, the word length Xn is “16”, when En = 1, the word length Xn is “16”, when En = 2, the word length Xn is “17”, and when En = 3, the word length Xn is “16”. When the length Xn is “20” and En = 4, the word length Xn changes to “27”.

  On the other hand, the internal processing word length (fixed-point format) of the signal processing apparatus is generally a multiple of 8, that is, 16 bits, 24 bits, or 32 bits, and the word length that matches the word length Xn is other than 16 bits. There is not. In other words, when converting fixed-point data to floating-point data, the word length of the fixed-point format with good conversion efficiency is limited. Therefore, if the word length of the external memory does not match the processing word length of the signal processing apparatus, the waveform data converted from the fixed-point format to the floating-point format is discarded, which leads to sound quality degradation.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide a waveform data converter capable of avoiding sound quality deterioration due to truncation while ensuring necessary resolution.

  In order to achieve the above object, according to the first aspect of the present invention, the number of powers necessary for expressing the fixed-point waveform data with the word length of the mantissa part is thinned out to a number that can be specified by the exponent part. An assigning unit for assigning the subtracted powers to the exponent part value, and a floating-point waveform data composed of an exponent part and a mantissa part according to the exponent part value assigned by the assigning means. Conversion means for converting to decimal point waveform data.

  In the invention according to claim 2, the fixed-point waveform data is composed of an exponent part having a first word length and a mantissa part having a second word length forming a word length corresponding to the storage width of the storage means. The floating-point waveform data is converted into floating-point waveform data, and the number of powers required to express the fixed-point waveform data by the mantissa part of the second word length is expressed by the exponent part of the first word length. Allocation means for decimating every other power until it becomes a specifiable number, and assigning the discontinuous number of powers thinned to the value specified by the exponent part of the first word length, and assigned by the allocation means Selection means for selecting an exponent value corresponding to fixed-point waveform data among exponent values, and fixed-point waveform data as the second word according to the exponent value selected by the selection means A conversion means for converting to a long mantissa part and selected by the selection means Output means for outputting floating-point waveform data composed of the exponent part of the first word length representing the value of the exponent part and the mantissa part of the second word length converted by the conversion means. It is characterized by that.

  In the invention according to claim 3, which is dependent on claim 2, the assigning means calculates the number of powers of odd columns thinned every other number until the number that can be specified by the exponent part of the first word length is reached. The first word length is assigned to a value designated by the exponent part.

  In the invention according to claim 4, which is dependent on claim 2, the assigning means calculates the number of powers of even-numbered columns obtained by thinning out every other number until the number that can be specified by the exponent part of the first word length is reached. The first word length is assigned to a value designated by the exponent part.

  According to the first aspect of the present invention, the number of powers required to express the fixed-point waveform data with the word length of the mantissa part is thinned out to a number that can be specified by the exponent part, and the discontinuity is thinned out. A large power is assigned to the exponent part value, and the fixed-point waveform data is converted into floating-point waveform data composed of an exponent part and a mantissa part according to the assigned exponent part value. Thereby, when the fixed-point waveform data is converted into the floating-point waveform data, it is possible to avoid deterioration in sound quality due to the truncation of the waveform data while ensuring the necessary resolution.

  According to the second aspect of the present invention, the fixed-point waveform data is obtained from the exponent part of the first word length and the mantissa part of the second word length forming the word length corresponding to the storage width of the storage means. When converting to floating-point waveform data, the number of powers required to represent fixed-point waveform data in the mantissa part of the second word length can be specified by the exponent part of the first word length. Thin out every other number until it reaches a certain number. The discontinuous powers thus thinned out are assigned to the value specified by the exponent part of the first word length, and the exponent part value corresponding to the fixed-point waveform data is selected from the assigned exponent part values. To do. Then, when the fixed-point waveform data is converted into the mantissa part of the second word length according to the value of the selected exponent part, the exponent part of the first word length representing the value of the selected exponent part and the conversion Since the floating point waveform data composed of the mantissa part of the second word length obtained by the above is output, it is possible to avoid deterioration in sound quality due to truncation of the waveform data while ensuring the necessary resolution.

(1) Outline of the Invention FIG. 1 is a diagram for explaining the outline of the present invention. As described above, when the waveform data in the fixed-point format is converted into the waveform data in the floating-point format, it is preferable that the word length width of the external memory matches the processing word length of the signal processing device. For example, assuming that the word length (storage width) of the external memory is 16 bits, a floating-point format with an exponent part of 3 bits and a mantissa part of 13 bits can express only up to 20 bits of fixed point. Therefore, if the processing word length of the signal processing device is 24 bits, the lower 4 bits are truncated. On the other hand, if the floating-point format has an exponent part of 4 bits and a mantissa part of 12 bits, it can support up to 27 bits of fixed point, but the resolution is halved.

  Therefore, in the waveform data conversion apparatus according to the present invention, as shown in FIG. 1, for example, waveform data of a fixed point of 24 bits can be efficiently stored in an external memory having a word length of 16 bits and an exponent part of 3 bits and a mantissa. The data is converted into floating point 16-bit waveform data composed of 13 bits. That is, in the floating-point 16-bit waveform data, the exponent of the upper 3 bits (15SB to 13SB) represents the power number. The number of powers that can be expressed by a 3-bit exponent is “0” to “7”. On the other hand, in order to express 24-bit data with the mantissa part 13 bits, power numbers from “0” to “11” are required.

  Therefore, it is not necessary to thin out “1”, “3”, “5”, and “7” from among the power numbers “0” to “11” necessary for expressing 24-bit data with the mantissa part 13 bits. The continuous power numbers “0”, “2”, “4”, “6”, “8”, “9”, “10”, “11” are replaced with values of exponent parts of 15SB to 13SB. In this way, the mantissa part 13 bits become significant digits because the original power numbers are “0”, “2”, “4”, “6”, “8”, “9”, “10” and “11”. Is the case. On the other hand, the thinned-out power numbers “1”, “3”, “5”, and “7” are the power numbers “2” and “4” that are necessary for expressing 24-bit data with a mantissa part of 13 bits. , “6”, “8”, and when these power numbers are used, they are converted to a floating point 16 bits consisting of a 4-bit exponent part and a 12-bit mantissa part.

(2) Embodiment Next, an embodiment of the present invention will be described with reference to FIGS. FIG. 2 is a block diagram illustrating a configuration of the signal processing device 20 according to the embodiment. A signal processing device 20 shown in this figure is used as a sound source of an electronic musical instrument, and includes a digital signal processor (hereinafter referred to as DSP) 21 and a data format converter 22. The DSP 21 performs a product-sum operation on, for example, fixed-point 24-bit waveform data under the control of a CPU (not shown). The data format converter 22 converts the fixed-point 24-bit waveform data supplied from the A / D converter 10 into floating-point 16-bit waveform data and stores it in the external memory 30 under the control of the DSP 21. . Further, under the control of the DSP 21, the data format converter 22 converts the floating point 16-bit waveform data read from the external memory 30 into the fixed-point 24-bit waveform data and outputs it to the DSP 21.

  The DSP 21 reads the waveform data stored in the external memory 30 via the data format converter 22 at every sampling period, performs, for example, a delay feedback operation on the read waveform data, and converts the musical sound data generated thereby to D / A output to the converter 40. The D / A converter 40 generates a tone signal obtained by D / A converting the tone data output from the DSP 21.

  Next, the configuration of the data format converter 22 will be described with reference to FIGS. The data format converter 22 converts a fixed-point 24-bit waveform data into floating-point 16-bit waveform data, and converts the floating-point 16-bit waveform data into fixed-point 24-bit waveform data. It is roughly divided into a second conversion unit.

  FIG. 3 is a block diagram illustrating a configuration of the first conversion unit. As shown in FIG. 1, the first conversion unit includes “1”, “3” among the power numbers “0” to “11” necessary for expressing the data of 24 bits in the mantissa part 13 bits. ”,“ 5 ”,“ 7 ”, and discontinuous power numbers“ 0 ”,“ 2 ”,“ 4 ”,“ 6 ”,“ 8 ”,“ 9 ”,“ 10 ”,“ 11 ”, While replacing with the value of the exponent part of 15SB to 13SB, according to the value of the replaced exponent part, the fixed-point 24-bit waveform data is expressed by the 13-bit mantissa part.

  Although not shown in FIG. 3, the number of powers to be thinned out by the replacement (“1”, “3”, “5”, “7”), that is, the original power numbers “2”, “4” ”,“ 6 ”,“ 8 ”, a configuration for converting fixed-point 24-bit waveform data into a floating-point 16-bit consisting of a 4-bit exponent and a 12-bit mantissa is provided. Good.

  The first conversion unit includes exclusive OR circuits 231 to 237, AND gate circuits 241 to 246, an inverting circuit 251, an 8-3 encoder 25, and an 8-1 selector 26. The exclusive OR circuit 231 is “H” level when all 12 bits from 23SB to 12SB out of 23SB (the most significant bit MSB) to 0SB (the least significant bit LSB) of the input fixed-point 24-bit data match. Generate output. Similarly, the exclusive OR circuit 232 generates an “H” level output when all 10 bits from 23SB to 14SB match. The exclusive OR circuit 233 generates an “H” level output when all 8 bits from 23SB to 16SB match.

  The exclusive OR circuit 234 generates an “H” level output when all 6 bits from 23SB to 18SB match. The exclusive OR circuit 235 generates an “H” level output when all four bits from 23SB to 20SB match. The exclusive OR circuit 236 generates an “H” level output when all three bits from 23SB to 21SB match. Exclusive OR circuit 237 generates an “H” level output when two bits from 23SB to 22SB match.

  The AND gate circuits 241 to 246 and the inverting circuit 251 prevent the exclusive OR circuits 231 to 237 from supplying a plurality of “H” level outputs to the 8-3 encoder 25 at the same time, and give priority to more matches. To do. The 8-3 encoder 25 generates an exponent value obtained by encoding the number of the input terminal supplied with the “H” level output from the previous stage among the input terminals “0” to “7” into 3 bits. . The 8-1 selector 26 selects 13-bit data supplied to the input terminal designated by the value of the 3-bit exponent generated by the 8-3 encoder 25 from the input terminals “0” to “7”. This is output as a mantissa part.

  In the 8-1 selector 26, 13 bits from 12SB to 0SB (LSB) of fixed-point 24-bit data are supplied to the input terminal “0”. Similarly, 13 bits from 14 SB to 2 SB are supplied to the input terminal “1”. The input terminal “21 is supplied with 13 bits from 16SB to 4SB. The input terminal“ 3 ”is supplied with 13 bits from 18SB to 6SB. 13 bits from 20SB to 8SB are supplied to the input terminal “4”. 13 bits from 21SB to 9SB are supplied to the input terminal “5”. The input terminal “6” is supplied with 13 bits from 22SB to 10SB. The input terminal “7” is supplied with 13 bits from 23SB to 11SB.

  According to the above configuration, the fixed-point 24-bit data is composed of a 3-bit exponent part output from the 8-3 encoder 25 and a 13-bit mantissa part output from the 8-1 selector 26. Converted to 16-bit data. The relationship between the converted floating-point 16-bit data and fixed-point 24-bit data is as shown in FIG.

  Next, FIG. 4 is a block diagram showing a configuration of the second conversion unit. The second conversion unit includes an 8-1 selector 27. The 8-1 selector 27 supplies to the input terminal designated by the value of the exponent part in the floating point 16-bit data (15SB to 13SB in the floating point 16-bit data) among the input terminals “0” to “7”. 24-bit data to be selected is output as fixed-point 24-bit data.

  The next 24-bit data is supplied to the input terminal “0” of the 8-1 selector 27. That is, 24-bit data obtained by bundling 12SB of 11 bits and 12SB to 0SB among floating-point 16-bit data consisting of 15SB (most significant bit MSB) to 0SB (least significant bit LSB) is supplied. The input terminal “1” is supplied with 24-bit data obtained by bundling 12SB for 9 bits, 12SB to 0SB, and 2 bits for “0”. The input terminal “2” is supplied with 24-bit data in which 12SB is bundled with 7 bits, 12SB to 0SB, and 4 bits “0” is bundled. The input terminal “3” is supplied with 24-bit data bundled with 5 bits of 12SB, 12SB to 0SB, and “0” of 6 bits.

  The input terminal “4” is supplied with 24-bit data obtained by bundling 3 bits of 12SB, 12SB to 0SB, and “0” of 9 bits. The input terminal “5” is supplied with 24-bit data obtained by bundling 2 bits of 12SB, 12SB to 0SB, and “0” of 9 bits. The input terminal “6” is supplied with 24-bit data obtained by bundling 12SB for 1 bit, 12SB to 0SB, and 10-bit “0”. The input terminal “7” is supplied with 24-bit data in which 12SB to 0SB and 11 bits of “0” are bundled.

  Next, the conversion accuracy of the data format converter 22 will be described with reference to FIG. FIG. 5 illustrates the characteristics of noise levels A to C when the signal level is linearly reduced to the dynamic range width of the fixed point 24-bit full range. The noise level A represents the noise level of floating point 16-bit data consisting of a 3-bit exponent and a 13-bit mantissa. Noise level B represents the noise level of floating point 16-bit data consisting of a 4-bit exponent and a 12-bit mantissa. The noise level C represents the noise level of floating point 16-bit data converted by the data format converter 22 and composed of a 3-bit exponent part and a 13-bit mantissa part to which discontinuous power numbers are assigned.

  As shown in FIG. 5, the noise level C has a resolution equivalent to that of the noise level A in a region where the signal level is high, while it is in the noise level B in a region where the signal level is close to the fixed point 24-bit full range. Since it has equivalent resolution, the necessary resolution can be ensured.

  As described above, the data format converter 22 converts the 24-bit waveform data in the fixed-point format into an unreadable format so that the processing word length of the signal processing device 20 can be efficiently stored in harmony with the word length of the external memory 30. While converting to the floating-point waveform data composed of a 3-bit exponent and a 13-bit mantissa for assigning consecutive power numbers, this 16-bit floating-point waveform data is converted to a 24-bit fixed-point format. Therefore, deterioration in sound quality due to truncation of waveform data can be avoided while ensuring the necessary resolution.

  In the above-described embodiment, “1”, “3”, “5”, “5”, among the power numbers “0” to “11” necessary for expressing the 24-bit length data with the mantissa part 13 bits, Replace the discontinuous power numbers “0”, “2”, “4”, “6”, “8”, “9”, “10”, and “11” with “7” thinned out with the value of the exponent part. However, instead of this, discontinuous power numbers “0”, “1”, “3”, “5”, “5”, “2”, “4”, “6”, “8” are thinned out. It is also possible to replace the “7”, “9”, “10”, and “11” with the value of the exponent part.

  Furthermore, the gist of the present invention is not limited to the conversion of fixed-point 24-bit data to floating-point 16-bit data. For example, as shown in FIG. 6, the mantissa part 21 bits represents 32-bit data. Among the numbers of powers “0” to “11” necessary for the discontinuous powers “0”, “2”, “4”, which are thinned out by thinning out “1”, “3”, “5”, “7”, By replacing “6”, “8”, “9”, “10”, and “11” with the value of the exponent part, it is also possible to convert fixed-point 32-bit data to floating-point 24-bit data. .

It is a figure for demonstrating the outline | summary of this invention. It is a block diagram which shows the structure of one Embodiment of this invention. It is a block diagram which shows the structure of the data format converter 22 which converts fixed point 24 bit data into floating point 16 bit data. It is a block diagram which shows the structure of the data format converter 22 which converts floating point 16 bit data into fixed point 24 bit data. It is a figure for demonstrating the conversion precision of the data format converter. It is a figure for demonstrating an example which converts fixed point 32-bit data into floating point 24-bit data. It is a figure for demonstrating an example which converts fixed point 20 bit data into floating point 16 bit data.

Explanation of symbols

10 A / D converter 20 Signal processing device 21 DSP
22 Data format converter 30 External memory 40 D / A converter

Claims (4)

Allocate the number of powers required to represent fixed-point waveform data in the mantissa word length to the number that can be specified in the exponent part, and assign the discontinuous powers that are thinned out to the exponent part value Means,
Waveform data comprising: conversion means for converting fixed-point waveform data into floating-point waveform data composed of an exponent part and a mantissa part according to the value of the exponent part assigned by the assigning means Conversion device. Apparatus for converting fixed-point waveform data into floating-point waveform data composed of an exponent part having a first word length and a mantissa part having a second word length that form a word length corresponding to the storage width of the storage means Because
Thinning out the number of powers necessary to represent the fixed-point waveform data in the mantissa part of the second word length until it becomes a number that can be specified by the exponent part of the first word length, Allocating means for allocating the discontinuous powers that are thinned out to a value specified by the exponent part of the first word length;
Selecting means for selecting an exponent part value corresponding to fixed-point waveform data among the exponent part values assigned by the assigning means;
Conversion means for converting fixed-point waveform data into the mantissa part of the second word length in accordance with the value of the exponent part selected by the selection means;
Floating point waveform data composed of the exponent part of the first word length representing the value of the exponent part selected by the selection means and the mantissa part of the second word length converted by the conversion means is output. And a waveform data converter characterized by comprising: The assigning means assigns the number of powers in the odd number sequence thinned out every other number until the number that can be designated by the exponent part of the first word length is assigned to the value designated by the exponent part of the first word length. 3. The waveform data conversion apparatus according to claim 2, wherein The assigning means assigns even numbers of powers of even columns thinned every other number until a number that can be designated by the exponent part of the first word length is assigned to a value designated by the exponent part of the first word length. 3. The waveform data conversion apparatus according to claim 2, wherein

Images