JPH02233031A - Block floating circuit - Google Patents

Abstract

PURPOSE:To record information onto a disk efficiently by detecting a polarity of an inputted digital signal or a determined amplitude, giving the result to an algorithm conversion, detecting a shift with an absolute value of a peak value subject to algorithm conversion and preventing the deviation of distribution of an error between input and output. CONSTITUTION:An inputted digital signal is divided into a period of a length being a time axis and when the quasi-momentary compression processing floating the in-period data with a shift detected from the peak in the period is implemented by using 2's complement, a polarity discriminator 32 detects the polarity of the inputted digital signal or the determined amplitude and an algorithm converter 33 uses an output of the polarity discriminator 32 to convert algorithm. Then a shift quantity detector 36 is used to detect the shift quantity with an absolute value 34 of the peak value subject to algorithm conversion thereby preventing the deviation of the distribution of input error.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to digital audio encoder technology.

[Conventional technology]

CD-ROM for the purpose of past reference in the civil and educational fields
As a disk-based computer system, there is a CD-1 that can handle three types of information: audio, images, and computer data.
1 unique bit compression technology is used, and this information can be efficiently recorded on a single disc.

The digitization of musical tones used in CDs is based on the PCM method using direct quantization with a sampling frequency of 44.1 KHz and an i-column bit number of 16 bits.

In CD-'I, as shown in Table 1, the normal 16
In addition to bit PCM, there are three types with different transmission bands and compression rates.
There are different levels available, which can be selected and used depending on the purpose of the disc program creator. (z) PCM is aimed at super high-fidelity and C
This is digital audio exactly the same as D.

(21 ADPCML/Be/I/A is intended for Hi-Fi, and is said to have sound quality equivalent to that of an LP. In the case of a music-only program, one disc can play for two hours in stereo.
It can be played in monaural for 4 hours.

(3) Level B is aimed at Mid-fi and has sound quality equivalent to FM. 4 hours in stereo, 8 hours in monaural
It is possible to regenerate time.

(4) Level C is intended for speeches,
It can play for 8 hours in stereo and 16 hours in monaural.

First, let us explain the surface body. FIG. 6 is a block diagram of the encoder 2I and decoder 22 of the CD-I standard audio bit compression system. The encoder 21 side includes an adaptive prediction filter unit 2
3, requantization section 24, and adaptive predictive quantization control section 25.

The adaptive prediction filter unit 23 is an FIR filter consisting of a predictor 1 and an adder 2, and is controlled by the control unit 3 to select the filter with the highest gain reduction effect among the four types of filters provided in the standard. be controlled. The 16-bit music signal input to the input 4 of the encoder 21 is delta-modulated by the adaptive predictive filter section 23.

In the requantization unit 24 at the next stage, the adder 5 first performs the ADPC
In the encoding algorithm of M, all lids 7,
The signal is shifted to the left (gain up) by the number of bits specified by the control unit 3 and inputted to the quantizer 8 . The quantizer 8 performs rounding processing to round off the data to the required number of bits, and outputs the result from the encoder output 9.

Adder 10, shifter 11, and predictor 6 constitute a noise shaper. The adder 10 calculates the error between the input and output of the quantizer 8 (that is, the quantization error), and the error is shifted to the right (gain down) by the same number of bits as the shifter 7 in the shifter l1 and sent to the predictor 6. is input. Predictor 6
Then, it is controlled by the control unit 3 so that it has the same transfer function as the predictor 1.

The adaptive prediction/quantization control unit 25 is the adaptive prediction filter unit 23
The selection of the prediction filter and the shift amount of each shifter 7, 1l corresponding to the quantization step are determined, and the settings are updated every 28 samples. The shift amount is determined by multiplying the peak value of the musical tone signal in 28 samples by 2'', and calculating the maximum value of 16 bits (
Forward direction 7FFF (lax). Negative direction 8000 (Hex
) ) is carried out by selecting the largest n that does not exceed Filter data indicating which filter was selected and range data indicating how many bits were shifted are transferred to the decoder 22 once every 28 samples.

The role of the noise shaper used in this encoding algorithm is to also whiten the requantization error that appears in the decoder output. Without a noise saver, the quantization error caused by the requantization section is
Since it can be regarded as white noise that is uniformly distributed between X (encoderization steps), the spectrum of the noise that appears in the decoder output depends on the frequency characteristics of the filter in the decoder section (i.e., the encoder.
PF2 and PF3 end up amplifying low-frequency noise.

Next, the principle of bit compression will be briefly explained. The encoder performs quasi-instantaneous compression by subjecting the residuals of the prediction filter to block floating processing every 28 samples.

Encoder human power x (n), prediction filter residual d (n
), quantization error e (n), encoder output d (n)
, decoder ^/ input d (n), decoder output x (n), and their respective Z transformations as X (Z), D (Z), E (Z).
D (Z). D (Z), X (Z), when there is no transmission error between the encoder output and decoder input, D
(Z) = G-X(Z) (1-P(Z)) +E(
Z) (1-R(Z)) -(1)X' (Z)
- G-'- D(Z)/ (1-P(Z))
−・−−−−T2)X' (Z)=X(Z)+
G-'- E(Z) (1-R(2)) / {1-
P(z))1・−・1・・(3) It becomes.

P (Z) is the transfer function of the predictor R (Z) is the transfer function of the noise shaper where R (Z
)·P (Z), the quantization error can be whitened. G is the gain (shift amount) when normalizing the residual of the prediction filter by the maximum value in the block, and G = 2"-"
(Level A), G-2"-" (Level B.C)
, and R is called range data. The transfer function of the prediction filter is expressed as H (Z) = 1 - P (Z), and the following four types are defined in the standard, and one is selected for each block depending on the spectral distribution of the input signal.

(l) PFI; H(Z)=1 (F=0) F=
FILTER DATA VALIIE (21 PF2
; H(Z)=1-0.93752 (F=1)(3)
PF3; H(Z)=1-1.796875Z+0
．． 8125Z (F=2) (41 PF3F H(Z
)=1-1.53125Z+0.859375Z (
F=3) The frequency characteristics of the four types of filters are shown in FIG. 5, and the block diagram of the adaptive predictive filter section is shown in FIG. PF2
, PF3 is designed for low-mid range, PF3 is designed for mid-high range, and PFl is designed for high range. Ideally, the prediction filter section of the encoder selects a filter such that the prediction residual within a block is minimized.

The simplest adaptation algorithm uses four filters and one block of memory l2a for each output.
-d and peak hold circuits 13a to 13d are prepared, the filter with the lowest peak value in a block is adopted as the filter of that block, and the data of the memory corresponding to 8 in the next block is output. Note that the filter data F is transmitted together with the range data R once for each data block before the audio data.

[Problem to be solved by the invention]

In the case of encoding, problems in realizing compression processing will be described using an example of a 16-bit system.

As a result of quantization having positive and negative asymmetry, the sign of the two's complement number is asymmetric, and the maximum positive value is smaller in absolute value than the maximum negative value by ILSB. If the number of bits is large, this effect is very small, but if compression processing is performed to reduce the number of bits, the weight of the ILSB increases and becomes unignorable.

The distribution range of the quantization error e (n) differs depending on the sign of the input signal. In the case of R(Z)・O, d (n)・d
(n) and the quantization step is Δq. When d(n)<0, -172・Δq<e(n)<+1/2・Δqd(n)
>0, -1/2. Δq<e(n)<+1/2. The Δq+b parameter b is determined by the level of d(n) and the number of quantization bits. For example, when setting it to 4 bits, 7 < d(n)
<F(hex) b. 0c(n)=G"-
If we take d(n)-d(n)=e' (n)-e(n) and take the Z transformation on both sides, we get C(Z)=E'(Z) { 1-R(Z)
}Thus, II! '(Z)・c(Z)/ { 1-R(
Z) } −−−−−−−(4) In the peak detection circuit shown in Figure 2, if the shift amount is simply determined from the peak value of the absolute value of the prediction filter output, the following will occur depending on the state of the input signal. A phenomenon is observed. When x(n) is a low-frequency signal and d(n) is continuously near the positive saturation level of the shifter, c(n) is (-G・ΔQ<c(n)<
0 ) and continues to have negative values, and c (n) contains many middle and low frequency components. When PF2 or PF3 is selected, equation (4) indicates mid-low range amplification, so e'
(n) increases, causing waveform distortion of the decoder output.
FIG. 4 shows how the output waveform is distorted when an IOOHZ sine wave is input under such conditions.

[Means to solve the problem]

The present invention utilizes quasi-instantaneous compression processing that divides an input digital signal into sections of a certain length on the time axis, and floats the data within the section using a shift amount detected from the peak value within the section using two-complement numbers. In this case, there is a detection means for detecting the polarity or fixed amplitude of the input digital signal, an algorithm conversion means for converting the algorithm using the output of the detection means, and a shift amount using the absolute value of the peak value converted by the algorithm. The block floating processing circuit includes a shift amount detector for detecting a shift amount and prevents bias in the distribution of errors between inputs.

〔Example〕

FIG. I is a block diagram showing one embodiment of the present invention, FIG. 2 is a block diagram of an adaptive predictive filter section, and FIG. 6 is a block diagram of an encoder and decoder. In the encoder 21, the adaptive predictive filter section 23 and the requantization section 24 are controlled by the adaptive predictive quantization control section 25, and the input from the digital input terminal 4 is compressed. The compressed encoder output 9 is reproduced by the decoder 22. The present invention relates to a floating circuit included in the adaptive prediction filter section 23.

When a digital input is input from the input 4 of the adaptive prediction filter unit 23 in FIG. 2, the filter characteristic P of the filter predictor l
The signals are filtered by FI, PF2, PF3, and PF4, stored in memories 12a-d, word-delayed for two days, and sent to requantizer 8.

The peak hold 13aNd has the configuration shown in FIG. 1, and each filter output is input to the manual input 3I, and if the polarity discriminator 32 determines that it is positive, the algorithm converter 33 is operated to change (a) to 1.1 or 1. If it is negative, it is multiplied by a number greater than or equal to 1 such as .2, and if it is negative, it is passed through to make the absolute value 34, the peak is held in the peak hold circuit 35, the shift amount is detected by the shift amount detector 36, and the adaptive predictive quantization control section Let the control scene be R. In FIG. 3, the polarity discriminator 32 of FIG. 1 is replaced with a constant detector 32', and the constant detector may be stored in the ROM table. When the detection by the constant detector is negative, 1 is set as a constant at the input of the algorithm converter 33', and when it is positive, the value of the constant is changed according to the amplitude value, and the saturation level of the chic is raised in the positive direction to correct the asymmetry. conduct.

〔Effect of the invention〕

According to the present invention, it is possible to perform correction in which the saturation level at the maximum value of ten or -peak values is made effective by one bit.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a block floating process showing an embodiment of the present invention, FIG. 2 is a block diagram of a filter using an embodiment of the present invention, and FIG. 3 is a block diagram of another embodiment of the present invention. 4 is a block diagram of the block floating process, FIG. 4 is a decoder output waveform diagram of the waveform distortion that the present invention aims to solve, FIG. 5 is a diagram showing the frequency characteristics of the filter, and FIG. 6 is a diagram showing the frequency characteristics of the filter.
The figure is a block diagram of an encoder and a decoder. 32...Polarity discriminator 33.33'...Algorithm converter 32'...Constant detector 34...Absolute value 35...Peak hold 36...Shift amount detector.

Claims (1)

[Claims] When performing quasi-instantaneous compression processing using two's complement that divides the input digital signal into sections of a certain length on the time axis and floats the data within the section using the shift amount detected from the peak value within the section. a detection means for detecting the polarity or fixed amplitude of an input digital signal; an algorithm conversion means for converting the algorithm using the output of the detection means; and a shift amount for detecting the shift amount from the absolute value of the peak value converted by the algorithm. 1. A block floating processing circuit comprising a detector and preventing bias in the distribution of errors between input and output.