JPS6063593A - Waveform generator for electronic musical instrument - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] Technical Field The present invention relates to a waveform generator for an electronic musical instrument, and particularly to a waveform generator for storing waveform data in a compressed representation format in a waveform memory.
This invention relates to reducing storage capacity and simplifying subsequent waveform calculations.

BACKGROUND OF THE INVENTION Conventionally, in waveform storage devices for electronic musical instruments, waveform data values are stored as they are in a linear or logarithmic representation format. Therefore, the number of bits of data increases,
There was a tendency for memory capacity to increase.

In particular, when attempting to obtain a high-quality musical waveform signal equivalent to that of a natural musical instrument by storing data of the entire waveform or multiple periods of waveforms from the start to the end of a musical tone in a memory and reading it out, 1 If the waveform data for the sample points has many pits, the overall memory capacity may become enormous. If the number of bits per piece of data is reduced in order to avoid this problem, the dynamic range will become narrower, resulting in a loss of sound quality.

In addition, if the waveform data is multi-bit, the arithmetic circuit for level control becomes large, and the digital-to-analog converter also becomes large, resulting in high cost.Objective of the InventionThe present invention eliminates the above-mentioned drawbacks. The purpose is to process digital waveform data in a data representation format that has a small number of bits but has sufficient dynamic range, and ultimately generates the desired waveform reliably. It is an object of the present invention to provide a waveform generator for an electronic musical instrument.

Summary of the Invention The feature of the present invention is to represent difference data of amplitude values between adjacent sample points regarding the valley sample point of a desired waveform in a floating point representation consisting of a mantissa part and an exponent part, and to store this in a waveform memory. That's the point. Then, by using a floating digital-to-analog converter to convert the mantissa data corresponding to the read output of the waveform memory into analog at a ratio according to the exponent data, the real value of the difference data for each sample point is converted into an analog signal. Melt. By cumulatively adding and subtracting the analog signal of the differential data determined in this way by an analog accumulator, an analog amplitude value for each sample point can be determined. By storing the difference data rather than the waveform amplitude value itself, the number of data bits per sample point can be reduced, and by representing the difference data in floating point representation, the number of data bits can be further reduced. . Further, when weighting calculation is performed on data read from the waveform memory, a floating point type calculation circuit can be used, so the configuration of the calculation circuit can be simplified.

Further, since cumulative addition and subtraction are performed after converting the floating point type difference data into a real-value analog signal, an analog 7-+ umulator with a simple configuration can be used as an arithmetic circuit for this purpose.

Embodiment In FIG. 1, the waveform memory 10 stores, for example, the difference in amplitude value between adjacent sample points for each sample point of a musical sound waveform consisting of multiple cycles from the start to the end of sound as shown in FIG. Data expressed in floating point representation is stored. An enlarged example of the waveform sample point amplitude value is shown in Figure 3, where sample points 0,
The amplitude value of 1,2°3.4... is A. g Al 1 A
2) If A3 p A4..., sample point 1
, 2, 3, 4... difference data △al, ΔFLts△
a3tΔa4... are respectively Δa1=AI AO, Δa2
=A2 A1 gΔa, =A, -A2, △a4 =
A4 A3, -. The waveform memory 10 stores differential data Δal, Δa2 corresponding to each sample point.
The digital value of ... (representatively expressed as Δai) is △
ai=M-Bl! + is decomposed and expressed in floating point representation, and the data of the mantissa part M and the data of the exponent part E are stored.

B is the base number, and in binary system, B=2 in general.

An example of the storage format of the waveform memory 10 is as shown in FIG. 4, in which data M of the mantissa part M of the difference data Δai determined for each sample point as described above. -Mn and data E of exponent part E. −En for each sample point address 0−
I remember each one in response to NK.

A keyboard 11 is used as a means for specifying the pitch of a musical tone to be generated, and information (key code KC) indicating the key pressed on the keyboard 11 is given to an address data generator 12. The address data generator 12 is a means for reading waveform data from the waveform memory 10 in accordance with the pitch specified on the keyboard 11, and reads sample point address data that sequentially changes at a rate corresponding to the specified pitch. Occur. The sample point address data generated from the address data generator 12 is applied to the address input of the waveform memory 10, and the mantissa data M corresponding to each sample point address O-N stored therein. -Mn and exponent data E. - Read out En sequentially. The address data generator 12 is reset to the initial address (ie, sample point address O) by a key-on pulse KONP that is generated instantaneously when a key on the keyboard 11 is pressed. Since any known technique can be used for the address data generation technique in the address data generator 12, its details will not be particularly shown.

The mantissa data M and the exponent data E read from the waveform memory 10 are sent to a multiplier 16 and an adder 14 for level control.
are input respectively. The key scaling or touch function generator 15 generates level coefficient data for key scaling according to a predetermined key scaling function using the key code KC of the pressed key as a parameter, or generates level coefficient data for touch response control by generating touch detection data. It is generated according to a predetermined touch function as a parameter, and this level coefficient data is k = m・2
8 (where k is a level coefficient) consisting of mantissa data m and exponent data e expressed in floating point representation.

The multiplier 16 multiplies the mantissa data M and m, and the adder 14 adds the exponent data E and e, resulting in a weighted operation at the real number level of ``k・△aIJ''. Perform the equivalent weighting operation at the floating point level, i.e. k・△ai=m・2°・M・21e: (m6M), (2
The first term (m-'M) on the right-most side is multiplied by the multiplier 16, and the exponent part of the second term (2e+'') is added by the adder 14.

By the way, looking at the sequence of mantissa data M, as shown in Table 1, the maximum value 11 corresponds to each value of exponent data E.
11... (all "1#") to minimum value 0000...
(All “0”) shows continuity, but the exponent data E
There is discontinuity between different values of . In other words, when the exponent part data E is "rO", the mantissa part data M directly corresponds to a real number, but when the exponent part data E is other than "0", the mantissa part data M does not directly correspond to a real number. As shown in the right column of Table 1, the mantissa data M with "1" added to the upper bit corresponds to the real number.
By doing so, the continuity of the sequence of mantissa data M is ensured between the exponent data E having different values.

Table 1 Therefore, in the example shown in FIG. 1, all bits of the exponent data E read from the waveform memory 10 are input to the OR circuit 16, and if E\0, the signal "1" is output from the OR circuit 16. death,
This is the mantissa data M read out from the waveform memory 10.
Add it as data corresponding to the MSB (most significant bit),
The signal is input to a multiplier 16. Similarly, all pits of the exponent part data e of the level coefficient are input to the OR circuit 17.

If e\0, the OR circuit 17 outputs a signal “1”, which is sent to the most significant bit MSH of the mantissa data m of the level coefficient.
is added as data corresponding to , and is input to the multiplier 16.

The mantissa data output from the multiplier 16 and the exponent data output from the adder 14 are floating (floating type).
The signal is input to a digital-to-analog converter 18, where it is converted from a floating point representation to a real number representing a sampling point amplitude value, and is also converted into an analog signal.

This floating digital to analog converter 18 is
A digital-to-analog converter 19 for converting mantissa data into an analog signal, and a variable voltage divider 20 for dividing (level shifting) the analog voltage of the converted mantissa data at a ratio according to the value of exponent data. Contains. In the variable voltage divider 20, when the base B of the floating point representation is "2" and the value of the exponent part data is the maximum value "7", the voltage division ratio is 2°=1.

The mantissa data in floating point representation is converted into analog data at a ratio according to the exponent data, and thereby the difference data △ai in floating point representation (more specifically, the difference data k・△ai weighted by the coefficients) is converted into analog data. Convert to real value.

The analog difference data output from the digital-to-analog converter 18 is input to the analog accumulator 21,
Additions and subtractions are cumulative. As a result, the accumulator 21 obtains an analog signal having a waveform amplitude value of Σ k ·Δai i ==0 for each valley sample point i. For example, at sample point 4 in FIG. 3, =k - A.

(However, △a, = Ao = O) is complete. The output of accumulator 21 is provided to a sound system 22.

In the above embodiment, it is assumed that the level coefficient does not change over time, but it may change over time. Also, the level coefficient k
may be expressed as a real number, and the mantissa data M may be multiplied by this.

In the above embodiment, the exponent part data E is stored in the waveform memory 0 for each sample point address, but it may be stored for each frame as follows. Divide the entire section of the waveform to be stored into multiple frames along the time axis, and perform this frame division and decomposition into floating point representation so that the value of exponent data E is common within each frame. . FIG. 2 also shows an example of frame division, in which the number of frames is 8, and the values of exponent part data E corresponding to frames 0 to 7 are "7" to "0", respectively. In this case, the waveform memory 10 consists of a mantissa data memory 10A and an exponent data memory IOB as shown in FIG. 5, and the mantissa data memory 10A has a mantissa corresponding to each sample point as shown in FIG. Department data M. -Mn are stored in the valley sample point addresses 0-H, respectively, and the exponent part data memory 10B has the exponent part data E=7 to 0 and the sample point address corresponding to each frame as shown in FIG. 6(b). Range data is stored in each frame address 0-7, respectively.

The storage format of the mantissa data memory 10A will be explained in detail with reference to FIG. 6(a).
The sample point address 0-G corresponding to the corresponding frame 0
Mantissa data M corresponding to each sample point. ~IVIg
are stored in the sample point addresses G+1 to H corresponding to frame 1, and mantissa data cho +1 to 2 corresponding to each sample point of frame 1 are stored, respectively, and as shown below, each frame 2 to Sample point addresses corresponding to 7 H+1~I, H+1~J, J+1~K, K
Mantissa data Mh+1 to Mn corresponding to the respective sample points are stored in 11 to L, L+1 to M, and M+1 to N, respectively. Here, GN is G<H<I<J<K
< L < M < N is any integer of the function.

rN+IJ corresponds to the total number of samples. And exponent data memo! In 710B, as shown in FIG. 6(b), exponent part data E=7 to 0 of the value corresponding to each frame is stored in each frame address O to 7 corresponding to frames O to 7, respectively. Each frame address in the memory 10B further stores data for specifying the range of sample point addresses corresponding to that frame. This sample point address range data includes, for example, the final sample point address G, H, I, J, L,
This is data showing IVf, N.

In FIG. 5, counter 23 and comparator 24 are for generating frame address data. The count output of the counter 23 is given as frame address data to the address input of the exponent data memory 10B, and the exponent part data E=7 to 0 corresponding to each frame address O to 7 stored therein and the sample point address range Read data (final address data) sequentially. The sample point address range data read from the memo January OB is applied to one input of the comparator 24, and the sample point address data from the address data generator 12 is applied to the other input. The comparator 24 outputs a signal “1” from the coincidence output EQ when the two forces match, and this is output to the counter 2.
6 count input. The counter 26 is reset to the initial address "0" at the start of sound generation by the key-on pulse KONP. Therefore, first, exponent part data E-7 and sample point address range data G corresponding to frame 0 from frame address O are read from the memo DB. The sample point address is 0.1°2.3...
When the final sample point address G in frame 0 is reached, the signal u 1 is output from the comparator 24.
71 is output, and the counter 26 is incremented by one. In this way, the frame address data is switched to "1", the data corresponding to frame 1 is read from the memory 10B, and the count contents of the counter 26 change until the sample point address reaches the final address H of frame 1. do not. In this way, while the sample point address data generated from the address data generator 12 belongs to the same frame, the contents of the counter 23 do not change, and the frame address data corresponding to that frame is output, and the sample point address The frame address data changes every time the frame to which the data belongs changes. Therefore, the combination of the mantissa data M (bio-Mn) of each sample point read from the mantissa data memory 10A and the exponent data E (7 to 0) read from the exponent data memory 10B at the same time. The real number of the waveform amplitude value of each sample point can be specified by In addition, counter 2
When the count output of 6 reaches the maximum value "7", the condition of the AND circuit 25 is satisfied and the counting operation of the counter 26 is stopped. If the exponent part data E is stored in correspondence with the frame address in this way, the waveform memory 1
The storage configuration of 0 can be further reduced.

In the above embodiment, the waveform memory 10 stores the amplitude value difference data regarding each sample point of the musical sound waveform during the entire sound generation period in floating point representation. It may also be stored in a format.

Further, the waveform generator of the present invention is not limited to the above-mentioned musical waveforms, but can be applied to the generation of envelope waveforms and other arbitrary waveforms.

Effects of the Invention As described above, according to the present invention, the difference data of amplitude values between each sample point is expressed in floating point representation, so the number of bits of waveform data per sample point can be significantly reduced. At the same time, a sufficient dynamic range can be secured. Therefore, the amount of storage in the waveform memory can be reduced, and the weighting arithmetic circuit can be simplified by using a floating point type arithmetic circuit, which is economical. Further, since the digital-to-analog converter may be of a floating type that inputs a small number of bits of digital data consisting of a mantissa part and an exponent part, the circuit configuration can be made smaller and lower in cost. Furthermore, since the differential data is accumulated by an analog accumulator, the structure of the accumulator is simplified.

[Brief explanation of drawings]

FIG. 1 is an electrical block diagram of the overall configuration of an electronic musical instrument showing one embodiment of the present invention, and FIG. 2 is a diagram showing a musical sound waveform over the entire sound generation period as an example of the waveform to be stored in the waveform memory in the same embodiment. FIG. 3 is an enlarged view of an example of the waveform sample point amplitude value, FIG. 4 is a view showing an example of the data storage format of the waveform memory in the same embodiment, and FIG. 5 is a block diagram showing an example of changing the waveform memory. FIG. 6 is a diagram illustrating -fPf- the data storage format of the waveform memory of FIG. 10... Waveform memory, 11... Conical disk, 12... Address data generator, 18... Floating digital to analog converter, 21... Analog accumulator, 13, 14... For weighting calculation multipliers and adders. Applicant: Nippon Musical Instruments Manufacturing Co., Ltd. Agent: Yoshihito Iizuka 1% AQ JP-A-Sho GO-63593 (7) 1 - Kizuna

Claims (1)

[Scope of Claims] 1. A waveform memory in which the difference data of amplitude values between adjacent sample points for each sample point of a desired waveform is represented in a floating point representation consisting of a coefficient part and an exponent part, and is stored in advance; A floating digital-to-analog converter converts the mantissa data corresponding to the read output of the waveform memory into analog at a ratio according to the exponent data; A waveform generator for an electronic musical instrument, comprising an analog accumulator that obtains an analog signal corresponding to a value. 2. The mantissa data and exponent data corresponding to the readout output of the waveform memory to be converted by the floating digital-to-analog converter are obtained by appropriately weighting the mantissa data and exponent data read from the waveform memory. A waveform generating device for an electronic musical instrument according to claim 1, which is data. 3. The weighting coefficients are expressed in a floating point representation consisting of a mantissa and an exponent, and the weighting is performed by multiplying the mantissa data and adding the exponent data. A waveform generating device for an electronic musical instrument according to claim 2. 4. The waveform generator for an electronic musical instrument according to claim 1, wherein the desired waveform is a musical sound waveform. 5. The waveform generator for an electronic musical instrument according to claim 1, wherein the desired waveform is an envelope waveform.