JP3256399B2 - Apparatus and method for detecting linearity error of digital audio signal, and apparatus and method for reducing linearity error - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and a method for detecting a linearity error of a digital audio signal having a distortion based on a linearity error generated during A / D conversion, and to an apparatus and a method for reducing the linearity error. is there.

[0002]

2. Description of the Related Art Conventionally, a linearity error generated in an A / D converter has been conventionally reduced by adjusting an internal resistance value of the A / D converter or adjusting a bit weight by an external circuit. . In the conventional method, the linearity error can be reduced at the time of A / D conversion. However, if the data is stored once with a distortion based on the linearity error at the time of the A / D conversion, the distortion cannot be reduced. Many music software programs that have been digitally recorded using an A / D converter in the past involve distortion based on a considerable linearity error due to poor conversion accuracy of the past A / D converter. But,
It has not been possible to reduce this with conventional methods.

On the other hand, as a device for detecting a linearity error generated in an A / D converter, a measuring device using a sine wave histogram method has been conventionally known. However, in this measuring device, a pure sine wave without distortion is converted into A
The linearity error is measured by inputting the signal to a / D converter and obtaining a histogram. Therefore, when a signal source that is not a pure sine wave without distortion such as a digital audio signal is used as a signal source, such a test apparatus cannot detect the linearity error itself.

[0004]

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and has been made in consideration of the above circumstances. It is an object to provide a reduction device and a reduction method. A linearity error detecting device and method suitable for the linearity error reducing device and the linearizing error reducing method, wherein the linearity error can be detected even when a pure sine wave without distortion is not used as a signal source. It is an object of the present invention to provide a sex error detection device and a detection method.

[0005]

SUMMARY OF THE INVENTION The present invention is a digital audio signal linearity error detecting apparatus for detecting the frequency of occurrence of each code from at least a part of a digital audio signal converted into a code by an A / D converter. By inputting the output of the occurrence frequency detection means and the occurrence frequency detection means, and comparing the appearance frequency of each code with the average appearance frequency of each code and the code whose quantization level is in the vicinity of each code. And a linearity error extracting means for detecting a code having a linearity error. The present invention also provides a device for reducing the linearity error of a digital audio signal.
An appearance frequency detection unit for detecting the appearance frequency of each code from at least a part of the digital audio signal converted into a code by the / D converter, and an output of the appearance frequency detection unit, and inputting the appearance frequency of each code, By comparing each code and its quantization level with the average frequency of occurrence in a code near each code, a linearity error extracting means for detecting a code having a linearity error, and a linearity error extracting means The digital audio signal converted into a code by the A / D converter, based on the detected code having the linearity error, has a linearity error reducing means for correcting the code having the linearity error. Is what you do.

The present invention also relates to a method for detecting a linearity error of a digital audio signal, wherein the frequency of occurrence of each code is detected from at least a part of the digital audio signal converted into a code by an A / D converter, and the detected code is detected. The code having a linearity error is detected by comparing the frequency of appearance of each code with the average frequency of appearance of each code and the code whose quantization level is in the vicinity of each code. . In the method for reducing a linearity error of a digital audio signal, a digital code converted by the A / D converter into a code based on a code having a linearity error detected by the method for detecting a linearity error of the digital audio signal. It is characterized in that a code having a linearity error in an audio signal is corrected.

[0007]

In the digital audio signal linearity error detecting apparatus and method according to the present invention, the frequency of occurrence of each code is detected from at least a part of the digital audio signal converted into a code by the A / D converter.
By comparing the frequency of occurrence of each code with the average frequency of occurrence of each code and the code whose quantization level is in the vicinity of each code, a code having a linearity error is detected, thereby obtaining the digital audio signal. Even if the A / D converter used for the A / D conversion is not provided, the linearity error of the digital audio signal depending on the A / D converter can be detected.

Further, according to the apparatus and method for reducing the linearity error of a digital audio signal of the present invention, the detected linearity error is generated using the apparatus and method for detecting a linearity error of the digital audio signal. A code used for A / D conversion of the digital audio signal is corrected by correcting a code having a linearity error in the digital audio signal converted into a code by the A / D converter based on the code. Even if the characteristics of the converter are unknown, the distortion due to the linearity error of the A / D converter can be reduced.

[0009]

FIG. 1 is a block diagram for explaining an outline of an embodiment of the present invention. In the figure, 1 is a switching means, 2 is a memory, 3 is a histogram forming means, 4 is a normalizing means, 5 is an error bit detecting means, and 6 is a digital signal processing means. In the first stage, a digital audio signal is read from a digital audio tape recorder (DAT), a compact disc (CD), or the like, and is input to the switching means 1 via a digital audio interface or the like. This digital audio signal is a signal obtained by converting an analog audio signal by an A / D converter before recording on a compact disc or the like. The read digital audio signal may be a part of a signal recorded on a digital audio tape recorder or the like. The switching means 1 outputs a digital audio signal to a memory 2 such as a hard disk, for example, and the memory 2 stores the digital audio signal.

Histogram forming means 3 reads the digital audio signal once stored in memory 2, detects the frequency of occurrence of each code representing a quantized amplitude level for a plurality of samples, and creates a histogram. I do. The normalizing means 4 includes the histogram generating means 3
, The linearity error of the A / D converter is extracted as a feature of the frequency of appearance, and is output as a normalized histogram. The error bit detecting means 5 receives the normalized histogram from the normalizing means 4, detects an error bit from the characteristic of the frequency of occurrence of the linearity error of the A / D converter, and converts the detection result into a digital signal processing means 6. Output to

In the second stage, a digital audio signal is read again from a digital audio tape recorder or the like, and is input to the switching means 1 via a digital audio interface or the like. The switching means 1 outputs a digital audio signal to the digital signal processing means 6. The digital signal processing means 6 receives the digital audio signal and outputs the error bit detection means 5.
A process of reducing the linearity error in real time by converting the audio signal converted to an erroneous code by the linearity error of the A / D converter into a value close to the original correct level based on the output of A digital audio signal whose distortion due to a linearity error has been corrected is output. The digital signal processing means 6 is realized by a digital signal processing processor (DSP) or the like.

In the above description, the digital audio signal is read again from the digital audio tape recorder or the like as the second stage. However, all the digital audio signals to be corrected for the distortion due to the linearity error are stored in the memory 2. In this case, the output of the memory 2 may be connected to the input of the digital signal processing means 6 via the switching means 1 or directly. When a large-capacity storage device including a semiconductor memory such as a RAM is used as the memory 2, the linearity error can be corrected in real time after receiving the output from the error bit detection unit 5 even in such a configuration. The digital audio signal thus output can be output.

When it is not necessary to output a digital audio signal whose linearity error has been corrected in real time, for example, when the output is to be stored in another storage device, a hard disk having a low data transfer rate is used as the memory 2. Alternatively, the output of the memory 2 may be connected to the input of the digital signal processing means 6 via the switching means 1 or directly.

When a digital audio signal is input from a storage device constituted by a semiconductor memory such as a ROM or a RAM and a digital audio signal having a corrected linearity error is output, the storage device itself is a memory 2. By controlling the storage device to achieve the above function, the output of this storage device can be directly connected to the inputs of the histogramming means 3 and the digital signal processing means 6.

Although the outline of the embodiment of the present invention is as described above, the function of each block shown in FIG. 1 will be described in detail below.

FIG. 2 is an example of a histogram of an actual digital audio signal. In the figure, the horizontal axis represents the code of the digital audio signal in decimal notation, and the vertical axis represents the frequency of appearance for each code. The code of the digital audio signal represents the amplitude level quantized by the A / D converter. In this histogram, the appearance frequency is high when the code is 0, and the appearance frequency gradually becomes gentler as the code increases in the positive direction and as the code decreases in the negative direction. The frequency of occurrence is substantially symmetrical in the positive and negative directions with the code being 0 as the center. Such a feature of the histogram is unique to each input audio signal. The illustrated histogram further includes a code whose appearance frequency sharply increases and a code whose sharpness decreases sharply as compared to adjacent codes. This peculiar appearance frequency is caused by a linearity error of the A / D converter.

The histogram forming means 3 shown in FIG.
The same device as the conventional sine wave histogram measuring device can be used. The digital audio signal is read out from the memory 2, the histogram memory is accessed using the code itself representing the quantized amplitude level as an address, and the data at this address is read out.
And write the added value to the same address. By performing this operation on multiple samples,
The cumulative frequency, that is, the frequency of appearance for each code is stored in the histogram memory. The plurality of samples need not be all of the digital audio signals for which the linearity error is to be corrected, but may be some of the samples converted by the same A / D converter. However, the accuracy of the histogram improves as the number of samples increases. As an example, when a sample of a period of 2,500,000 samples in which a relatively high level portion is continuous is used, a favorable histogram distribution can be obtained. Note that the histogram conversion can be realized by a digital counter circuit or software.

The normalizing means 4 shown in FIG. 1 receives the output of the histogram generating means 3 and compares the linearity error of the A / D converter with the frequency of occurrence by comparing it with the frequency of occurrence of the input audio signal. Is extracted as a feature. The average appearance frequency of each code and the code whose quantization level is in the vicinity of each code represents the appearance frequency unique to the input audio signal. Therefore, as an example, by dividing the frequency of appearance for each code by this average frequency of appearance, the linearity error of the A / D converter can be extracted as a feature of the frequency of appearance. More specifically, all codes are divided into a plurality of blocks with a plurality of codes having adjacent quantized amplitude levels, for example, four codes as one unit. The average frequency of appearance for each block is determined from the frequency of appearance of each code constituting this block.

A value obtained by dividing the frequency of appearance of each code constituting the block by the average frequency of appearance of each block is defined as the normalized frequency of appearance A of each code. That is,
(Frequency of each code divided into four) ÷
(Average value of the appearance frequencies of four codes) = A. The histogram consisting of the normalized appearance frequency is A / D
The feature of the appearance frequency due to the linearity error of the converter is extracted. The normalized appearance frequency A is obtained by substantially eliminating the characteristic of the appearance frequency unique to the audio signal. If there is no linearity error in the A / D converter,
The normalized appearance frequency A is 1 in all codes. As the average appearance frequency, a moving average of each appearance frequency of each code and a predetermined number of codes whose quantization levels are in the vicinity of each code may be used.

As a special case, a digital audio signal read from a digital audio tape recorder or the like may include a pure sine wave signal or a digitized tone signal similar thereto.
When such a signal exists, a normalized appearance frequency can be obtained by the same method as the calculation of the differential linearity error in the conventional sine wave histogram measuring device. A histogram of a pure sine wave without distortion is theoretically obtained, and is small when the amplitude level is 0, becomes gradually symmetrical in the positive and negative directions, and becomes maximum at the positive and negative peak levels. It becomes.
The theoretically obtained histogram represents the frequency of appearance of the input audio signal. Inputting a pure sine wave signal without distortion or a digital audio signal of a tone signal similar to this, and dividing the frequency of occurrence of each code by the frequency of occurrence of the same code obtained from the theoretical value of the sine wave histogram As a result, a normalized appearance frequency can be obtained. The value obtained by subtracting 1 from this value is the differential linearity error.

The error bit detecting means 5 receives the normalized histogram output from the normalizing means 4 and detects error bits. The normalized histogram includes a code having an extremely large number of occurrences and a code having an extremely small number of occurrences. In other words, it indicates the digit position of the bit.
The cause of the linearity error depends on the type of A / D converter when the audio signal is digitized. Here, a weight resistance type D / A converter or R-2
A successive approximation type A / D converter using a binary weight current source type D / A converter having an R ladder as an internal A / D converter is assumed. That is, assuming that the weight is used for setting the amplitude level as the threshold,
Focusing on the influence of a certain weight on multiple quantized amplitude levels, conversely, detecting the weight that caused a linearity error from multiple quantized amplitude levels, that is, the frequency of code occurrence It is to do.

[0022] For example, the spacing between the occurrence frequency is extremely large code 2, i.e., when the interval between the quantization levels is 2 ¹ times the quantization level of the least significant bit, error in the least significant bit You can see that there is. The interval is 2 ^3, that is, when the interval of the quantization level is 2 ³ times the quantization level of the least significant bit, that there is an error from the least significant bit to the bit digit position of the third bit Understand. When the interval between the quantization levels of a plurality of codes having extremely large numbers of appearances is 2 ⁿ , it can be seen that there is an error in the bit at the n-th digit position from the least significant bit. A threshold is required to determine that the frequency of appearance is extremely large. For example, if the normalized appearance frequency A is 1.2 or more, it can be determined that the appearance frequency is extremely large. This threshold is 1.5
It may be about. Conversely, the digit position of a bit having an error can be detected from the interval between codes having extremely small appearance frequencies. Alternatively, both can be detected in combination.

FIG. 3 is a flowchart for explaining processing steps for realizing the function of the error bit detecting means. S
At 11, the normalized histogram is input from the normalizing means 4. In S12, the initial value of the variable n is set to 1 based on the number of quantization bits of the digital audio signal.
Is subtracted. In S13, the normalized appearance frequency of each code is extracted for every 2 ⁿ quantization levels of each code. In S14, it is determined whether or not all the extracted normalized appearance frequencies are 1.2 or more. If it is 1.2 or more, the process proceeds to S15, and
If less, the process proceeds to S16. In S15,
Since it is known that there is an error at the nth bit from the least significant bit, the value of n is stored in the error bit memory. In S16, 1 is subtracted from the value of n to update the value of n. S1
At step 7, if the value of n is 0, the process proceeds to S18,
If the value of n is not 0, the process returns to S13. In S18, the error bits stored in the error bit memory are transferred to the digital signal processing means 6, and the process is terminated.

The digital signal processing means 6 receives a digital audio signal and performs processing for reducing a linearity error in real time based on the output of the error bit detection means 5. By converting some of a plurality of samples having a level whose appearance frequency increases due to an error into a level estimated to be an original correct level, distortion due to a linearity error of the A / D converter is reduced.

FIG. 4 is a flowchart showing a procedure for correcting a linearity error. In S21, the code of each sample of the digital audio signal to be corrected is input in real time. In S22, it is determined whether or not the value at the digit position of the error bit is 1 for each code of one sample. The code whose value at the digit position of the error bit is 1 is a code whose quantization level is affected by the error bit. If the value at the digit position of the error bit is 1, the process proceeds to S23, where 1
If not, the process proceeds to S24. In S23, the correct quantization level of the sample, that is,
The code is estimated, and the process proceeds to S24. In S24, the code of the input sample or S23
Is written to an internal file and output to the outside. At this time, the position on the time axis,
That is, data indicating the order of the sample from the beginning is also stored in the file together with the code.

LSB is the quantization level of the least significant bit, a and b are positive integers, and N is 0 or a positive integer. When the resistance value inside the successive approximation type A / D converter has an error of + b · LSB in the ath bit from the lower bit, the frequency of occurrence of the code having the quantization level of 2 ^a · N + 2 ^a−1 −1 is b + 1 And the quantization level is 2 ^a ·
The appearance frequency of each code from N + 2 ^a -1 to 2 ^a · N + 2 ^a -b becomes zero. Also, when there is an error of −b · LSB in the a bit from the lower bit, the frequency of occurrence of the quantization level of 2 ^a · N + 2 ^a −1 increases by b + 1 times,
The quantization level is 2 ^a · N + 2 ^a−1 −1 to 2 ^a · N + 2
^The frequency of occurrence of each code up to ^a-1 + b-1 is zero.

Therefore, if such an error is corrected so as to be canceled, the distortion due to the linearity error of the A / D converter is reduced. However, correctly, sample data that should be a code with a low frequency of occurrence is erroneously converted into a code with a high frequency of occurrence. Therefore, a plurality of samples having a quantization level with a large number of appearances include a sample that should originally have another quantization level.

In order to correct this, it is first necessary to check how much error conversion has occurred. A plurality of codes whose appearance frequency becomes extremely large under the influence of one error bit detected by the error bit detection means 5 can be theoretically obtained. Let X be the average of the frequency of occurrence of such multiple codes actually detected. Then, for a sample having a code whose occurrence frequency becomes extremely large under the influence of the error bit, the code is changed to a code whose occurrence frequency is theoretically extremely small by the one error bit.
Or, according to the value of the X, a plurality of codes, for example,
What is necessary is just to convert at a rate of one to X pieces. When there are errors in a plurality of bits, the above operation is performed for each of them.

The amplitude level quantized under the influence of the error bit detected by the error bit detection means 5 is:
In S23 shown in FIG. 4, it is determined which of positive and negative should be corrected based on the difference between the quantization levels of adjacent samples on the time axis.

FIG. 5 is an A / D conversion characteristic diagram for explaining a technique for estimating a correct code. In the figure, the horizontal axis represents analog input, and the vertical axis represents digital output after A / D conversion. Note that the broken line indicates the ideal characteristics of the A / D converter. For a code whose value at the digit position of the error bit is 1, the internal file is read to check how many samples the code continues on the time axis. For example, when it is determined that there is an error of b = + 1 · LSB in the a = 3th bit from the lower order, the input / output characteristics of the A / D converter are as shown in FIG. That is, 8
At N + 3, the digital output does not change with an increase in the analog input corresponding to + 1 · LSB. Therefore, 8
In N + 3, the frequency of appearance becomes extremely large. At 8N + 6, the digital output becomes 8N + 7 instead of 8N + 7 for an increase of the analog input corresponding to + 1 · LSB.
N + 8. In the characteristic diagram shown in FIG. 5, 8N + 3 to 8N
N + 6 is described.

FIG. 6 is an explanatory diagram showing a digital output code on a time axis for explaining a method of estimating a correct code. In the figure, the horizontal axis represents time, and the vertical axis represents the digital output code of the sample at each sampling time by the amplitude level. In order to reduce the linearity error, all digital outputs corresponding to 8N + 4 to 8N + 6 are converted to values larger by + 1 · LSB. Then, in FIG. 5, if the digital output samples corresponding to 8N + 3 are K ₁ , K ₂ ,..., K _t ,
When two or more samples corresponding to 8N + 3 continue for t samples or more, K ₁ and a sample K ₀ immediately before the sample K ₀
And the difference between the digital outputs is referred to as D. D
Is positive, the value from _{Kt / 2} to _Kt is incremented by 1, and D
Is negative, the value of K ₁ to K _{t / 2} is incremented by one.

By doing so, the digital output sample corresponding to 8N + 3 is converted to a value larger by + 1 · LSB once in two times. Which sample is to be transformed is determined by D, which is the difference between the quantization levels of adjacent samples on the time axis. In the above description, the case where it is determined that there is an error of b = + 1 · LSB has been described. However, even if the value of b is another value, a correct code can be similarly estimated. When the value of t is an odd number, the value is rounded down or rounded up so that the value of t / 2 becomes an integer. By the above operation, A
Data converted to an incorrect value due to the linearity error of the / D converter can be easily corrected, and the linearity error can be reduced.

As a method of correcting the linearity error, various interpolation techniques used when estimating the quantization level of a certain sample from the quantization levels of adjacent samples on the time axis can be used. That is, continuous samples having a code in which the error bit determined by the error bit detection unit 5 affects the quantization level can be corrected using these interpolation techniques.

In the above description, the embodiment as the linearity error reducing device of the A / D converter has been described.
Normalization means 4 used as an example of a part of the configuration
The above blocks can be used as a device for detecting the linearity error of the A / D converter.

[0035]

As is apparent from the above description, according to the present invention, even if the characteristics of the used A / D converter are not known, the music software having the distortion based on the linearity error of the A / D converter. There is an effect that the distortion of the music can be reduced and the quality of the music software can be improved. Even if the used A / D converter itself is not used,
There is an effect that a linearity error of the / D converter can be detected.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an outline of an embodiment of the present invention.

FIG. 2 is an example of a histogram of an actual digital audio signal.

FIG. 3 is a flowchart illustrating processing steps for realizing a function of an error bit detection unit.

FIG. 4 is a flowchart showing a procedure for correcting a linearity error on a time axis.

FIG. 5 is an A / D conversion characteristic diagram for explaining a method of estimating a correct code.

FIG. 6 is an explanatory diagram showing a digital output code on a time axis for explaining a technique for estimating a correct code.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Switching means, 2 ... Memory, 3 ... Histogram conversion means, 4 ... Normalization means, 5 ... Error bit detection means, 6 ... Digital signal processing means.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-59-127419 (JP, A) JP-A-4-144423 (JP, A) JP-A-5-333140 (JP, A) 300854 (JP, A) JP-A-64-78525 (JP, A) JP-A-62-72226 (JP, A) JP-A-4-31834 (JP, U) JP-A-3-113538 (JP, U) (58) Field surveyed (Int. Cl. ⁷ , DB name) H03M 1/00-1/88

Claims (9)

(57) [Claims] 1. An appearance frequency detection means for detecting an appearance frequency of each code from at least a part of a digital audio signal converted into a code by an A / D converter, and an output of the appearance frequency detection means being input, and A linearity error extracting means for detecting a code having a linearity error by comparing the frequency of occurrence of each code with the average frequency of occurrence of each code and the code whose quantization level is in the vicinity of each code. An apparatus for detecting a linearity error of a digital audio signal. 2. The linearity error extracting means normalizes the frequency of occurrence of each detected code by an average frequency of occurrence of each code and a code whose quantization level is in the vicinity of each code. 2. The apparatus for detecting a linearity error of a digital audio signal according to claim 1, wherein the apparatus performs comparison. 3. An appearance frequency detection means for detecting an appearance frequency of each code from at least a part of a digital audio signal converted into a code by an A / D converter, and an output of the appearance frequency detection means being inputted, and A linearity error extracting means for detecting a code having a linearity error by comparing the frequency of occurrence of each code with the average frequency of occurrence of each code and a code whose quantization level is in the vicinity of each code; Based on the code having the linearity error detected by the linearity error extracting means, the A / D
An apparatus for reducing a linearity error of a digital audio signal, comprising: a linearity error reducing means for correcting a code having a linearity error in a digital audio signal converted into a code by a converter. 4. The linearity error extracting means normalizes the frequency of occurrence of each detected code by an average frequency of occurrence of each code and a code whose quantization level is in the vicinity of each code. 4. The apparatus according to claim 3, wherein the comparison is performed. 5. The linearity error reducing unit detects a bit in which an error has occurred in a code in which the linearity error has occurred from an interval between quantization levels of the code in which the linearity error has occurred, and corrects the bit. The apparatus for reducing a linearity error of a digital audio signal according to claim 3 or 4, wherein: 6. An appearance frequency of each code is detected from at least a part of the digital audio signal converted into a code by the A / D converter, and the appearance frequency of each detected code is determined by each code and its quantization level. A method for detecting a linearity error in a digital audio signal, wherein a code having a linearity error is detected by comparing the average appearance frequency of a code near each code. 7. The appearance frequency of each detected code is represented by:
7. The method for detecting a linearity error of a digital audio signal according to claim 6, wherein the comparison is performed by normalizing each code and its quantization level by an average frequency of occurrence in a code near each code. . 8. A digital audio signal converted by the A / D converter into a code based on a code having a linearity error detected by the digital audio signal linearity error detection method according to claim 6 or 7. A method for reducing a linearity error of a digital audio signal, comprising correcting a code having a linearity error in the signal. 9. The method according to claim 1, wherein a bit having an error in the code having the linearity error is detected based on an interval between quantization levels of the code having the linearity error, and the bit is corrected. 9. The method for reducing a linearity error of a digital audio signal according to item 8.