JP3408140B2 - Information encoding method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information encoding method, and more particularly to an information encoding method for encoding and writing information on a recording medium such as MD or DCC.

[0002]

2. Description of the Related Art FIG. 10 is a block diagram showing the structure of a mini disk system. The main configuration of the mini disk system will be described with reference to FIG. The mini disk 1 is housed in the cartridge 2, and the signal read from the optical pickup 3 is input to the RF amplifier 4 and becomes an RF signal during reproduction. AD embedded in disc
The IP signal is decoded by the address decoder 5,
Used for servo control. RF signal is EFM / C
The IRC / encoder / decoder 6, the shock proof memory controller 7 and the audio compression ATRAC encoder / decoder 8 perform digital processing, and the D / A converter 9 converts the digital processing into an analog signal to output an audio signal.

On the other hand, at the time of recording, the audio input signal is quantized by the A / D converter 10 and digitally processed in the route opposite to that at the time of reproduction. After that, the recording head 12 is driven by the head drive circuit 11 so that the audio signal is magnetically modulated and recorded on the disk.

FIG. 11 shows the voice compression ATRA shown in FIG.
3 is a block diagram showing an encoder portion of a C encoder / decoder 8. FIG. In FIG. 11, ATRAC (Ad
aptive TRansform Acoustic Coding) encoder is M
This is a high-efficiency compression method adopted in the D system, and is a method in which time-axis data is converted into frequency-axis data, that is, spectrum data by orthogonal transformation, and the spectrum data is distributed and bit-allocated.

The reproduced audio signal is input to a band division filter (QMF) 81 and divided into two. The divided audio signal of one band is a band division filter 8
It is further divided into two bands by 2, and the original audio signal is divided into three bands of low band, middle band and high band. The audio signal of the other band divided by the band division filter 81 is delayed by the delay amount of the band division filter 82.
Delayed by 3.

The divided audio signal is MDCT8
4,85,86 perform the improved discrete cosine transform to transform the time series data into frequency spectrum data.
The converted frequency spectrum data of each band is input to the bit allocation + quantization circuit 90.

The audio signal divided into the above-mentioned three bands is input to block size selection circuits 87, 88 and 89, the block size for each band is determined and given to the psychoacoustic analysis circuit 91. The psychoacoustic analysis circuit 91 extracts the non-audible data, and compresses the transmission amount by selectively deleting the non-audible data when the audio data is quantized by the bit allocation + quantization circuit 90.

The bit allocation + quantization circuit 90 allocates the spectrum data of each band based on the auditory analysis processing of the psychoacoustic analysis circuit 91, performs normalization with the optimum number of bits in each processing bandwidth, and performs formatting. It is determined and output to the multiplexing circuit 92. Further, the psychoacoustic analysis circuit 91 includes MDCTs 84, 85,
From 86, spectral data of three bands of low band, middle band, and high band are given. Then, the psychoacoustic analysis circuit 91 gives the word length, which is the number of quantization bits, and the scale factor, which is the normalization scale, to the multiplexing circuit 92. The multiplexing circuit 92 outputs the multiplexed bit stream.

FIG. 12 shows the voice compression ATRA shown in FIG.
It is a figure which shows the relationship between the band division by C encoder / decoder, and MDCT. The audio data of 0 to fs / 2 is 0 to 0 by the band division filters 81 and 82 in FIG.
Low range of fs / 8, mid range of fs / 8 to fs / 4, fs
Divided into high frequencies from / 4 to fs / 2, 128/1 each
The spectrum is 28/256, which is represented as 512 samples in the entire band. Therefore, the frequency resolution is 43 Hz.

The spectrum data is divided into 256/32, 128/32 and 128/32 by the block size selection circuits 87, 88 and 89, as shown in FIG.

FIG. 13 is a diagram showing a time-series audio signal input to the ATRAC encoder, and FIG. 14 is a diagram showing that the audio signal shown in FIG. 13 is band-divided by the QMFs 81 and 82 of FIG. 11 and converted by the MDCTs 84 to 86. It is a figure which shows the frequency spectrum data produced, and the number of spectra is roughly shown. FIG. 15 is a diagram showing a signal level in which the amount of energy is obtained from spectrum data by the bit allocation + quantization circuit 90, and FIG. 16 is a diagram for explaining masking calculation.

The psychoacoustic analysis circuit 91 performs a psychoacoustic analysis process. Auditory psychology takes into account masking effects and minimum audible characteristics. The masking effect means that an unrecognizable area occurs because the spectrum shown by the diagonal lines around the spectrum of a certain loud sound is masked by the loud sound as shown in FIG. The total masking amount shown in FIG. 17 is determined by calculating the sum of the levels of the spectrum shown in FIG. 16A affecting other spectra from the signal level shown in FIG. 15 and performing masking calculation.

FIG. 18 is a diagram showing a minimum audible characteristic curve. The minimum audible characteristic is a characteristic indicating that a sound below the minimum audible limit cannot be heard, and the sensitivity is highest around 4 kHz, and the sensitivity is low in the high and low frequencies. Therefore, 1k
There is no problem even if the SN ratio is increased in the band of Hz to 5 kHz and the SN ratio is lowered in the low band and the high band compared to that.

Therefore, the bit allocation + quantization circuit 90 converts the signal level shown in FIG.
The total mask amount shown in and the spectrum below the minimum audible limit due to the maximum audible characteristic shown in FIG. 18 are thinned out. The state is shown in FIG.

The ratio between the signal level and the masked level shown in FIG. 19 is called SMR (Signal to Mask Ratio). In order to normalize this SMR, a scale factor serving as a scale for normalization and a word length that is the number of quantization bits are given to the multiplexing circuit 92 from the psychoacoustic analysis circuit 91.

The number of bits of the word length when the SMR is dB is predetermined, and conventionally, for example, when the SMR is 12 to 18 dB, the word length is assigned to 2 bits, and when the SMR is 6 dB or less, 0 bit is assigned. It was being done.

[0017]

As described above, since the effect of compressing the amount of information at the time of allocation is great, S
When 0 bit is assigned to a signal of very low level of MR, spectrum data is lost,
The continuity of the spectrum in the frequency direction is lost,
There was a drawback that it caused a problem in sound quality.

Therefore, a main object of the present invention is to provide an information coding method capable of ensuring the continuity of spectrum in the frequency direction, in view of the above problems.

[0019]

The invention according to claim 1 is
The analog data is converted into frequency data, the converted spectrum data within the predetermined processing bandwidth is converted into energy to calculate the signal level, and the processing band is calculated according to the relative ratio of the signal level to the predetermined masking level. Is an information encoding method for allocating the number of quantized bits in accordance with a predetermined linear rule, and according to the linear rule, even for a relative ratio area in which the number of allocated bits is zero,
It is characterized in that the minimum number of quantization bits assigned according to the linear rule is forcibly assigned.

According to a second aspect of the present invention, in the first aspect, the relative ratio area for forcibly allocating the number of quantization bits is up to the middle of the relative ratio and the zero decibel which is the limit by the linear rule. It is characterized by

According to a third aspect of the present invention, in the first or second aspect, the analog data is voice data, and the masking level is set according to how a person hears. .

According to a fourth aspect of the present invention, in the third aspect, the coded data is recorded on a mini disc, and the masking is a masking effect caused by a correlation between spectra and a human ear. It is characterized by being combined with a minimum audible characteristic.

[0023]

[0024]

[0025]

1 is a flow chart for explaining the operation of an embodiment of the present invention, and FIG.
5 is a flowchart for explaining a transmission amount adjustment operation in FIG.

The processes according to the flowcharts shown in FIGS. 1 and 2 are executed by the bit allocation + quantization circuit 90 shown in FIG. That is, the bit allocation + quantization circuit 90 causes S
The MR is normalized to determine the optimum transmission amount. Also,
The signal level is calculated from each spectrum data shown in FIG. In this signal level calculation, assuming that the spectrum data is S, the spectrum is obtained by the following equation, and thus the signal level represented by the energy amount is obtained as shown in FIG.

ΣS * S Next, the masking level is calculated. The masking level can be calculated by calculating the sum of the levels that each spectrum exerts on other spectra as shown in FIG. Specifically, the sum of the masking level of the critical band and the level of masking from other critical bands in the critical band unit is calculated, and the masking amount shown in FIG. 17 is calculated.

The minimum audible characteristic is preset. Then, the total masking calculation is performed by the minimum audible characteristic and the calculated masking amount, and the SMR is calculated by subtracting the total masking level from the signal level. That is, the white part shown in FIG. 19 is obtained. Bit allocation is performed on the obtained SMR.

FIG. 3 is a diagram showing the relationship between the word length and the number of quantization bits, and FIG. 4 is a diagram showing the relationship between the SMR and the word length.

In the bit allocation, as shown in FIGS. 3 and 4, the number of bits of the word length WL is assigned according to how many dB the SMR is, and the transmission amount is calculated.

Conventionally, when the SMR is allocated to the word length as described above, the word length is allocated to 2 bits when the SMR is 12 to 18 dB and 0 when the SMR is 12 dB or less.
Bits were allocated. This is because the minimum allocated bit is 2 bits because the bit representing the sign of ± is required, and therefore the minimum SMR is up to 12 dB according to the linear rule of the bit allocation shown in FIG. .
In this way, since the effect of compressing the amount of information at the time of allocation is large, if 0 bit is allocated to a signal having a very low SMR level, the spectrum data will be lost, and the continuity of spectrum in the frequency direction and The necessary overtone component is lost, causing a problem in sound quality.

Therefore, in one embodiment of the present invention, FIG.
As shown in (2), 2 bits are allocated even to a signal with a low level such as SMR of 6 dB to 12 dB, and the continuity of the spectrum in the frequency direction and the overtone component are ensured to improve the sound quality. However, if the quantized bits are allocated until the SMR becomes close to zero, unwanted noise sounds are picked up,
Or, the sound balance may be deteriorated, which may result in an offensive sound. According to the experiment, if the quantized bits are assigned up to SMR of about 6, the effect of improving the sound quality is great, but if the SMR smaller than this is assigned, noise sound becomes a concern. If the above-mentioned transmission amount calculation is not appropriate, the bisection method is used to correct the SMR.

FIG. 5 shows an SMR adjustment tree, and FIGS. 6 and 7 show how the transmission amount is brought close to the optimum value by the bisection method. In this embodiment, the adjustment value of the first step is set to 32 dB so that the adjustment of 128 dB is possible, and a step of ± 0.5 dB is provided for fine adjustment. The value at the time of this fine adjustment is obtained from the accuracy of parameters and the like. FIG. 6 shows an example where the transmission amount is adjusted by the bisection method, and the example where the transmission amount is converged at the 7th step is shown in FIG.

As shown in the SMR adjustment tree of FIG.
The number of steps to be divided into 2 from 2 dB to 0.5 dB is 7 steps. As shown in FIG. 6, after the processing of the seventh step,
If the transmission amount is less than or equal to the adjustment amount, the adjustment can be ended, but as shown in FIG. 7, the transmission amount may not be equal to or less than the value after the seventh step. In this case, another correction step is provided for adjustment.

In this embodiment, the adjustment is performed using the same value as in the seventh step which is the final adjustment step. When the transmission amount is adjusted by dividing the adjustment value, the transmission amount approaches the optimum transmission amount, but there is no guarantee that the transmission amount will be equal to or less than the optimum transmission amount. In order to ensure that the adjustment result of the transmission amount is less than or equal to the optimal transmission amount, the value is set to the same value as the final adjustment step value, and the transmission amount is forcibly reduced to or less than the optimal transmission amount.

The SMR adjustment is performed according to the procedure shown in FIG. That is, first, the transmission amount adjustment value Δ is set to the start value Δ = 32 dB according to the SMR adjustment tree shown in FIG. The transmission amount is calculated by the bit allocation method described above, and it is determined whether the transmission amount is larger or smaller than the optimum transmission amount. 32d from each band of SMR if many
B is subtracted, and if less, 32 dB is added to each band of SMR. Next, the Δ value is set to 16 dB, which is ½ of 32 dB, and the transmission amount is obtained from the SMR after addition and subtraction. It is again determined whether the transmission amount is larger or smaller than the optimum transmission amount, and if it is larger, 16 dB is subtracted from each band of SMR, and if it is smaller, 16 dB is added to each band of SMR.

This process is repeated, and Δ is 8, 4, 2,
The operation is sequentially performed by dividing into two such as 1 and 0.5 dB. In this case, if the transmission amount becomes less than or equal to the optimum transmission amount at the 7th step as shown in FIG. 6, the adjustment of the transmission amount is finished as it is, but if it becomes less than or equal to the optimum transmission amount at the 7th step as shown in FIG. If not, the value of Δ is set so as to be less than or equal to the optimum transmission amount in the eighth step, and the transmission amount is forcibly made less than or equal to the optimum value.

When the transmission amount is adjusted, the SMR amount is added or subtracted with the adjustment value Δ as a boundary line. The transition of this boundary line is shown in FIGS. FIG. 8 shows the case where the transmission amount becomes less than or equal to the optimum transmission amount at the seventh step, and FIG. 9 shows the case where the transmission amount does not become less than or equal to the optimum transmission amount at the seventh step.

If it is desired to complete the transmission amount adjustment calculation in 7 or 8 steps or less, the following method is also available. That is, when adjusting the transmission amount, the optimum transmission amount is changed to the optimum transmission amount −δ.
If it is in the range up to, terminate it immediately. At this time,
The value of δ is an arbitrary value set in advance. For example, in this embodiment, good results are obtained when δ = 0.5, which is the range of fine adjustment.

[0040]

As described above, according to the present invention, the spectrum level within the predetermined processing bandwidth is converted into energy to calculate the signal level, and the signal level is processed according to the relative ratio of the signal level to the predetermined masking level. When the number of quantized bits in a band is allocated, the continuity of the spectrum in the frequency direction can be ensured by increasing the width of the relative ratio corresponding to the same allocated bit in the region where the relative ratio is small. Also, when allocating according to the linear rule, the continuity of the spectrum in the frequency direction is determined by forcibly allocating the quantization bit number even in the relative ratio area where the allocated bit number is zero according to the linear rule. Can be secured.

[Brief description of drawings]

FIG. 1 is a flow chart for explaining the operation of an embodiment of the present invention.

FIG. 2 is a flowchart for explaining a transmission amount adjustment operation in FIG.

FIG. 3 is a diagram showing a relationship between a word length and the number of quantization bits.

FIG. 4 is a diagram showing a relationship between SMR and word length.

FIG. 5 is a diagram showing an SMR adjustment pattern.

FIG. 6 is a diagram showing an example of convergence in 7 steps by the bisection method.

FIG. 7 is a diagram showing an example in which the transmission amount is exceeded by the dichotomy and forced convergence is performed in 8 steps.

FIG. 8 is a transition diagram when the transmission amount becomes less than or equal to the optimum transmission amount in the seventh step.

FIG. 9 is a transition diagram when the transmission amount does not fall below the optimum transmission amount in the seventh step.

FIG. 10 is a block diagram showing a configuration of a mini disk system.

11 is a block diagram showing the audio compression ATRAC encoder shown in FIG.

12 is a diagram showing a relationship between band division and MDCT by the audio compression ATRAC encoder shown in FIG.

FIG. 13 is a diagram showing a time-series audio signal input to an ATRAC encoder.

FIG. 14 is a band division of the audio signal shown in FIG.
It is a figure which shows the frequency spectrum data converted by CT.

FIG. 15 is a diagram showing a signal level in which the amount of energy is obtained from spectrum data.

FIG. 16 is a diagram for explaining masking calculation.

FIG. 17 is a diagram showing a total mask amount.

FIG. 18 is a diagram showing a minimum audible characteristic curve.

FIG. 19 is a diagram in which the signal level shown in FIG. 15, the total mask amount of FIG. 17, and the minimum audible characteristic curve of FIG. 18 are superimposed.

[Explanation of symbols]

81,82 QMF 83 delay 84,85,86 MDCT 87,88,89 Block size selection circuit 90-bit allocation + quantization circuit 91 Auditory psychological analysis circuit 92 Multiplexing circuit

─────────────────────────────────────────────────── ─── Continuation of the front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) H03M 7/30 G11B 20/10 311

Claims (4)

(57) [Claims] 1. Converting analog data into frequency data
Then, the spectrum within this converted predetermined processing bandwidth
The signal level is calculated by converting the data into energy and
Phase of the signal level against a constant masking level
Predetermines the number of quantization bits in the processing band according to the comparison
Is an information encoding method for allocating according to the linear rule, and the relative ratio at which the number of allocated bits is zero according to the linear rule.
Even for regions, the maximum assigned according to the linear rule
Characterized by forcibly assigning a small number of quantization bits
Information encoding method. 2. The number of quantization bits is forcibly assigned
The relative ratio area is defined as the relative ratio and
Claim, characterized in that it is up to the middle of the rodecibel
The information encoding method according to Item 1. 3. The analog data is voice data.
The masking level is set according to how people hear
Claim 1 or a contract, characterized in that
The information encoding method according to claim 2. 4. The encoded data is a mini disc
Is recorded in the
Masking effect caused by correlation and minimum audibility of human ear
Claim, characterized in that it is a combination of characteristics and
3. The information encoding method described in 3.