JPS62101127A - Adaptive quantization device - Google Patents

Abstract

PURPOSE:To minimize the quantization distortion by applying a proper robust adaptive quantization independently on an input signal level through the control of a leakage coefficient by the input signal level. CONSTITUTION:The input signal d(k) inputted from an input terminal 1 is subject to logarithmic conversion at first by a logarithmic converter 2 into log2d(k), the result is subject to scaling by a subtractor 3 by an adaptive parameter y(k) to obtain Log2d(k)-y(k), the result is quantized by quantizer 4 and goes to a quantized output value I(k), which is outputted from an output terminal 5. An averaging circuit 8 supplies arithmetic means <y>(k) of the y(k) at a sampling section NT, a leakage coefficient controller 9 supplies reciprocal of output <y>(k) of the circuit 8 to obtain a leakage coefficient, then the result is subtract from a value 1 to output 1-beta(k). A multiplier 10 uses the value and an output from a delay device 7 to output (1-beta(k)y(k)), which is inputted to an adder 11 and a limiter 12 limits the variation width of the value y(k).

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention is an adaptive differential PCM (hereinafter referred to as ADPCM).
The present invention relates to a robust adaptive quantizer used for encoding, etc., and particularly to an adaptive quantizer suitable for quantizing signals with small fluctuations such as voice band data and tone signals.

[Background of the invention]

The robust adaptive quantizer is published in Japanese Patent Publication No. 59-500077.
No. 59-70329, IE
EE TRANSACTIONS ON COMMUNICATIONS
COMMUNCATIONS) November 1975 issue 13
It is disclosed on page 62, and has also been recommended as G721 by the International Telephone Consultative Committee CCITT.

The adaptive parameter of the robust adaptive quantizer increases or decreases depending on the level of the human signal, and the dynamic range of the human signal (signal after adaptation) of the quantizer increases.

Adapt to fit within the range of the quantizer. Since this adaptive parameter moves in proportion to the input level, the amount of leakage due to the fixed leakage coefficient depends on the human input signal level.

On the other hand, the adaptation amount for updating the adaptive parameters is determined by the quantizer output, but since the quantizer is designed to minimize the nesting error for a certain human signal distribution, the optimal case is The distribution of the quantizer output signal matches the distribution of the human signal. Therefore, the expected value for this distribution of the amount of adaptation (update) when the resonant distortion is the smallest is a certain constant amount and does not depend on the level of the human input signal.

An equilibrium state occurs when the leakage amount and this adaptation amount are equal, and the adaptation parameters are adapted around a certain equilibrium value, so one depends on the input level and the other does not depend on the input level. This fact means that the quantizer's manual signal distribution (distribution of the signal after adaptation) in the equilibrium state deviates from the optimal distribution depending on the input signal level, and quantization distortion increases. In other words, the quantization distortion of the conventional robust adaptive quantizer depends on the input signal level, and there is a risk of overload in some cases.

[Purpose of the invention]

An object of the present invention is to provide a robust adaptive quantizer that has good quantization characteristics regardless of the level of an input signal.

[Summary of the invention]

The problem with conventional robust adaptive quantizers is that the leakage coefficient is fixed regardless of the input signal level. Therefore, in the robust adaptive quantizer of the present invention, the adaptive parameters are averaged over a certain interval, and the leak coefficient is controlled by the average value, so that the leakage amount of the adaptive parameters does not depend on the level of the input signal. When the level is high, the leak coefficient is made small, and when the level is small, the leak coefficient is made large.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of an adaptive quantizer, with input terminals 1. Logarithmic converter 2. Subtractor 3. Quantizer 4. Output terminal 5. Adaptive power generator 6. Delay device 7. Averaging circuit 8. Leak coefficient controller 91 multiplier 10. It consists of an adder 11 and a limiter 12. Input signal d(k) input from input terminal 1 (k indicates sampling processing time kT).

T is the sampling time) is first logarithmically transformed by the logarithmic converter 2 to become Logtd(k), and then scaled by the subtracter 3 using the adaptive parameter y(k) to become Logl
d(k) y (k), which is further quantized by the quantizer 4 and outputted from the output terminal 5 as a quantized output value I(k).

The adaptive bin generator 6 adaptively j
d L ogt M (1) and adder 1
Add to 1.

As shown in FIG. 2, the averaging circuit 8 includes N-1 delay devices 13, an adder 14, and a multiplier 15.

Arithmetic mean value of y(k) in a certain sampling interval NT<
y>(k) is determined by the following equation (1).

The leak coefficient controller 9 includes an inverter I as shown in FIG.
6, a multiplier 17, and a subtracter 18.
6, take the reciprocal of the output <y>(k) of the averaging circuit 8 and get β
By multiplying by , the leakage coefficient; β (k) = β
(< y >(k))-'...(2) is calculated,
Next, by subtracting this from the value 1, 1-β(k) is output. Here, β is a certain constant (0<β×1), and y is the equilibrium center value of the adaptive parameter y (k) when the quantizer 4 is performing optimal quantization. For example, if the quantizer 4 is a 4-bit quantizer and is optimally designed for a certain Gaussian distribution, its distribution density function is as follows6 (mean=0.variance=,?) Here, as shown in Fig. 4, the quantizer performs quantization when X≧O, and performs quantization (symmetrical with respect to the origin) when x < O. do. If the probability of the quantized output value 1(k) at this time is P(i) (i = o ~ 7); P (i) = (probability that II(k)l = i)
...(4) Represented, 9 is; Σ P (i, i) LogM (i)==βta
...(5) Determined by i=0.

When 1-β(k) is output from the leak coefficient controller 9, 1
The multiplier 10 calculates (1-β) from this value and the output value of the delay device 7.
(k) y (k)) is output and input to the adder 11. As a result, the adder 11 performs the calculation of the following equation (6).

y (k+1)= (1-β(k) y (k) +
L Ogz M (i) (6) Note that the limiter 12 limits the range of variation in the value of y (k).

Now, let <y>(k)=a. In such a case, this is true at a certain input level when the input signal is a signal with small fluctuations, such as audio band data or a tone signal. In this case, β(k)=β according to equation (2), so by substituting this into equation (6), βy (k)” L og2M
(i). Therefore, the expected value β < y (k)> of the leakage amount of the adaptive parameter in the equilibrium state is; β < y (k)> = < L Ogz M (i)
>=ΣP (i) ・Log, M (i)i=Q. Based on this and equation (5); β< y (k)> = βta...(7
) holds true. That is, when the input signal has a Gaussian distribution as shown by the solid line I in FIG. 5, P(i)=P(i), #), and the quantizer 4 performs optimal quantization. Note that P (i) has a probability that I(k)=i.

However, suppose that a human signal larger than this level comes in, and consider its equilibrium state. That is, let us assume that <y>(k)>a. At this time, the leak sensitivity in the equilibrium state is β
If (k)=β, the expected value is β<y> (k)=
It is expressed as (1-β)<y>(k). On the other hand, the 6th
) According to Eq.

β<y>(k)=<LogM(i)>=Σ P (
i) Log M (i) i=0 〉Σ P (i, i) LogM (i)...(
8) i:.

becomes. In other words, (P(i))2≠(P(i, 2))'.

1=O1=0, which means that the quantizer 4 is not performing optimal quantization.

In general, (LogM(i))- is 1=0 0>LogM(0)<LogM(1)<・=<LogM
(7)>O−(9), so <y>(k
) > Input signal to childizer 4 (signal after adaptation)
The distribution becomes as indicated by the broken line H in FIG. 5, and an equilibrium state deviates from the optimal case is reached, making overload likely to occur. Also in the case of <y>(k)<, the state deviates from the optimum state and the quantization distortion increases due to the same principle as above.

However, in this embodiment, the leakage coefficient β(k) is
), the expected value of the leakage amount in an equilibrium state is β(k) y (k)> = <βp (< y >
(k)-' y (k)>, and in the case of a voice band signal with small fluctuations, <y>(k) is almost constant, so <β(k) y (k)> = β・＜
y >(k)-1< y (k)>=βri=ΣP
(i) LogM (i)i:0. In other words, it becomes independent of the input level.

In the case of voice band data, it is known that the short-time distribution is approximately Gaussian-shaped. That is, the quantizer 4 performs quantization in an optimal state.

〔Effect of the invention〕

According to the present invention, since the leak coefficient is controlled by the human signal level, optimal robust adaptive quantization can be performed regardless of the human signal level, which has the effect of minimizing quantization distortion. .

[Brief explanation of drawings]

Fig. 1 is a block diagram of an adaptive quantizer according to an embodiment of the present invention, Fig. 2 is a block diagram of an averaging circuit, Fig. 3 is a block diagram of a leak coefficient controller, and Fig. 4 is a block diagram of a quantizer. The characteristic graph of FIG. 5 is a distribution graph of the human input signal. Input terminal, 2: Logarithmic converter, 3: Subtractor, 4: ;
il quantizer, 5: output terminal, 6: adaptive quantity generator, 7:
Delay device, 8: Averaging circuit, 9: Leak coefficient controller, lO
Square multiplier, 11: Adder, 12: Limiter, 13: Delay device, 14: Adder, 15: Multiplier, 16: Inverter, 17
: Multiplier, 18: Subtractor. Agent Patent Attorney Tadashi Akimoto Figure 1 Figure 2

Claims (1)

[Claims] In a robust adaptive quantizer that quantizes a digitized input signal that is input at each sampling time, there is a means for taking the average value of an adaptive parameter in a certain sample period, and a leak coefficient of the adaptive parameter for making it robust. An adaptive quantizer comprising: means for controlling based on an average value.