JP2000057241A - Nonlinear arithmetic circuit - Google Patents

Abstract

PROBLEM TO BE SOLVED: To form an arbitrary nonlinear nonmotoneous transfer function generating circuit and an arbitrary nonlinear dynamics actualizing circuit which have simple circuit constitution and low power consumption. SOLUTION: A voltage source 3 which varies in voltage value with an arbitrary function of time including nonlinearity is connected to a capacitor 5 through a switch 4, which is turned on and off under the control of an input signal. The input signal is a pulse modulated signal having information when the voltage value or current value varies and the switch 4 is turned on the moment the voltage value or current value of the input signal varies to save the current value of the voltage source 3 in the capacitor 5. Then the terminal voltage across the capacitor is outputted and then the input signal is converted with a converting function of the same style with an arbitrary function and outputted.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a non-linear operation circuit, and more particularly to a circuit having a non-linear transfer characteristic suitable for integration into a circuit or a circuit for realizing non-linear analog dynamics.

[0002]

2. Description of the Related Art In recent years, there has been a remarkable improvement in the performance of computers. However, processing that can be easily performed by humans, such as recognizing human faces and reading facial expressions in real time, can hardly be executed by current computers. . This is because the processing method of the computer itself is not suitable for such processing.

Therefore, an intelligent processing model that mimics the information processing mode of the brain has been studied. A typical example is a neural network. Until the early 1990's, most neural network models were simple enough to obtain a weighted sum of inputs and perform threshold processing.
In recent years, many models characterized by non-linear processing, that is, non-monotonic processing more than threshold processing, have been proposed, and their high performance has been demonstrated. Examples include non-monotonic transfer characteristics,
That is, an associative memory in which units having non-monotonic input / output characteristics are connected to each other, or a non-monotonic dynamics, that is, a neural network that uses chaotic behavior generated by a time evolution equation for association and optimization processing.

In the first place, since the neural network model is a massively parallel / distributed information processing model, it is extremely inefficient to execute on a current Neumann computer which is a sequential processing method. For practical use of neural networks, it is essential to use integrated circuits as dedicated hardware. Therefore, in response to the development of the above model,
Various circuits have been proposed that realize nonlinear and non-monotonic characteristics and are suitable for forming an intelligent processing model that requires nonlinear characteristics.

On the other hand, in various information processing, a random number is often required. In many cases, a chaotic signal can be used as a random number. Therefore, many chaos generation circuits have been proposed. Since chaos can be basically generated by a feedback system having non-monotonic transfer characteristics, if a non-monotonic transfer function generating circuit with a small circuit scale and low power consumption can be made, a high-performance information processing circuit can be configured.

Conventionally, as a circuit for generating a non-linear transfer function, a characteristic peculiar to elements constituting the circuit has been used, or a circuit designed to obtain a specific transfer characteristic has been used. For this reason, there is a disadvantage that the function shape of the transfer function cannot be arbitrarily changed. That is, it is difficult to generate an arbitrary non-linear / non-monotonic transfer function using a normal analog circuit.

[0007] Therefore, as a circuit for generating an arbitrary transfer function, a method using a pulse width modulation (PWM) method is known as TM.
orie, S.Sakabayashi, M.Nagata, and A.Iwata, “Nonlinea
r function generators and chaotic signal generator
s using a pulse-width modulation method, ”Electron
ics Letters, vol. 33, no. 16, pp. 1351-1352, 1997 (literature
[1]).

To generate a PWM signal from an input voltage,
The pulse width is determined by comparison with a reference voltage waveform for comparison. By making the reference voltage waveform non-linear, non-linear conversion can be realized. That is, the reference voltage waveform is f
When expressed as a function of time (t), the input voltage is V ₁
Then, the pulse width obtained is F (V ₁ ).
Here, F = f ⁻¹ , that is, an inverse function of f. When a constant current source is switched with the obtained PWM waveform and charge is stored in a capacitor connected in series with the current, the terminal voltage V _{2 of the} capacitor becomes V ₂ = f ^-1 (V ₁ ) = F (V ₁ ) and a non-linear conversion (transfer function) can be realized. According to this principle, f (t) must be monotonic. Therefore, in order to generate a non-monotonic transfer function, comparators are prepared by the number of monotonic sections, and a logical operation of their outputs is performed to obtain a non-monotonic function. Produces a PWM signal having a pulse width according to

FIG. 10 shows an example of a conventional nonlinear function generation method. FIG. 10A shows an example of a circuit configuration in the case where the number of monotone sections is 2, and two comparators 601 and 602
Is prepared, and a PWM signal is generated by the AND gate 9. FIG.
(B) shows the change over time of the _two reference voltages f ₁ and f ₂ . The obtained PWM signal turns on the switch 4 for the duration of the pulse width, connects the constant current source 39 to the capacitor 4, and charges the capacitor 4 with a constant current. In this way, a charge proportional to the pulse width of the PWM signal is stored in the capacitor 4, and as a result, a voltage V ₂ proportional to the pulse width is generated at the output terminal 2. FIG. 10C shows the relationship between the input voltage V ₁ and the output voltage V ₂ converted by the non-linear function generation circuit of FIG. 10A using the reference voltages f ₁ and f ₂ .

The reference voltage waveform as an arbitrary function can be generated, for example, by a combination of a digital memory and a D / A converter. With a specific non-linear function shape, it is not so difficult to generate a function of time with an analog circuit. That is, the point of this circuit is to convert an arbitrary function of time into a transfer function. Since any given voltage waveform can be used in common by a plurality of transfer function generation circuits, only one voltage waveform generation circuit may be used even when many transfer function generation circuits are arranged on a chip and operated in parallel.

As an example, consider a "logistic mapping" represented by F (x) = 4ax (1-x). Here, a is a constant. Considering a discrete time system, the result of the logistic mapping at time t is fed back to the input,
Compute the logistic mapping of +1. That is, x
It is known that when a discrete-time dynamics of (t + 1) = 4ax (t) (1-x (t)) is executed, chaos occurs from a point where a slightly exceeds 0.89. A diagram in which output x is overwritten with a as a variable is called a branch diagram. FIG. 11 shows a bifurcation diagram of the logistic mapping obtained by the numerical calculation.

To realize this mapping by a circuit, FIG.
As shown in (b)

(Equation 1)

The two non-linear reference waveforms represented by ## EQU1 ## may be applied to comparators 601 and 602, respectively. Here, if the output is fed back to the input through the sample and hold circuit, the same dynamics can be executed.

FIG. 12 shows another conventional example. Comparator 60
When the components 1 and 602 are constituted by capacitors and inverters, the sample and hold function can be incorporated in the comparator. Further, the output terminal of FIG. 10A is connected to an input terminal via a buffer to form a feedback type. FIG. 10B shows the timings of the pulses φ ₁ , φ ₂ , φ ₃ and the temporal changes of the reference voltage waveforms f ₁ , f ₂ .

FIG. 13 shows a branch diagram generated by circuit simulation of this circuit. The characteristics similar to those of FIG. 11 are obtained, but the details are slightly blurred. This is due to a calculation error of the simulation.

[0016]

The above-described conventional nonlinear operation circuit has the following disadvantages.

1) It is necessary to prepare as many comparators as the number of monotonous sections of the nonlinear transfer function to be realized, which increases the circuit scale and power consumption.

2) A set of voltage waveforms that is the inverse of the nonlinear transfer function to be realized must be prepared, which complicates the control circuit.

3) Since a plurality of comparators and a logical operation circuit are combined, an operation error increases. The effect can be seen in FIG. 13, where the details are somewhat blurred.

Now, the pulse modulation system represented by PWM itself is also a method suitable for a large-scale integrated system for performing an analog operation. A. Iwata and M. Nagata,
“AConcept of Analog-Digital Merged Circuit Archit
ecture for Future VLSI's, ”IEICE Transactions on F
undamentals of Electronics, Communications and Comp
Utre Sciences, vol. E79-A, no. 2, pp. 145-157, 1996 (reference [2]). Conventional integrated circuits are classified into a digital system and an analog system, and each of them has advantages and disadvantages. The pulse modulation method is an excellent method that makes up for the disadvantages of both and combines the advantages of both. The pulse modulation method is particularly suitable for a large-scale analog processing circuit.

The digital system is resistant to noise and crosstalk between wirings, has good controllability, and is easily connected to ordinary computer systems because the same digital system is employed. The calculation accuracy can be increased as much as the number of calculation bits is increased. The size of the transistor used can be reduced year by year as the miniaturization of the LSI manufacturing process progresses.

However, the disadvantage of the digital system is that
The circuit scale is about two orders of magnitude larger than that of the analog system, and it is difficult to integrate a large number of units on a chip and perform parallel operations. For this reason, usually, time division processing is performed by one unit, and a large number of calculations are processed. Therefore, there is a disadvantage that the total processing time is long. Since the calculation accuracy, that is, the number of bits and the circuit scale are in inverse proportion, the circuit scale increases as the calculation accuracy increases. In a non-linear analog operation represented by chaos, a rounding error in digital calculation has a great influence, so that a considerably high precision is required, and as a result, a circuit scale becomes large. Therefore, it is difficult to execute the non-linear analog dynamics by a large-scale parallel operation by a digital method.

On the other hand, since the analog system has a small circuit scale, a large number of arithmetic units can be mounted on a chip, and parallel arithmetic can be easily executed. It is suitable for executing a neural network model using analog dynamics. However, contrary to the digital system, there are disadvantages that the system is easily affected by noise and crosstalk between wirings, it is difficult to obtain high calculation accuracy, and it is not easy to connect to other digital systems. Further, since the analog characteristics of the circuit elements are used, the dimensions of the elements cannot be reduced freely, and it is difficult to reduce the power consumption and the voltage. There is also a drawback that the benefit of miniaturization of the LSI manufacturing process cannot be obtained.

The signal used in the pulse modulation method is a pulse, has a binary voltage or current value, and has analog information such as a pulse width and a pulse phase in the time axis direction. Like the digital system, it has strong resistance to the influence of noise and the like, has good controllability, and is easily connected to a digital system. The calculation accuracy is usually about 6 to 8 bits, and has an accuracy capable of performing intelligent processing such as image / voice recognition. It should be noted that the pulse modulation method can process information in an analog manner, so that even with 6-bit precision, a quantization error does not occur as in the digital method. Therefore, it can be effectively applied to an analog neural network model in which continuity of information is important or a model using chaos. A circuit for implementing the pulse modulation method is mainly a switch or a capacitor, and does not use the analog characteristics of the element. Therefore, like the digital method, the element can be reduced in accordance with the miniaturization of the manufacturing process. The circuit scale is much smaller than that of the digital system, and is at the same level as that of the analog system. Furthermore, since one data is basically represented by one pulse, there is only one state transition, and the number of normal state transitions is smaller than that of a digital system in which data is represented by a number of bit state transitions. Can be considerably reduced.

Although the pulse modulation system has many advantages as described above, in order to construct an intelligent processing system based on large-scale nonlinear analog processing, a pulse modulation system capable of efficiently performing nonlinear calculations is required. Conventionally, FIG.
And only a nonlinear operation circuit as shown in FIG. 12 has been proposed.

Accordingly, it is an object of the present invention to provide an arbitrary nonlinear / non-monotonic transfer function generating circuit and an arbitrary nonlinear dynamics realizing circuit which overcome the above-mentioned drawbacks of the conventional example and have a simple circuit configuration and save power. With the goal. It is assumed that this circuit can handle not only a voltage signal but also a pulse modulation signal. Accordingly, it is an object of the present invention to realize an integrated circuit as hardware having a high intelligent processing capability in a field of image / voice recognition, robot / system control, prediction, and the like, which mimics a biological information processing mode.

It is another object of the present invention to provide a compact and highly integrated random number generation circuit applicable to various information processing circuits.

[0028]

According to the present invention, a voltage source whose voltage value changes as a function of an arbitrary time including non-linearity is connected to a capacitor via a switch, and the switch comprises:
Conduction and non-conduction are controlled by the input signal, and the input signal is a pulse modulation signal having information when the voltage value or the current value changes, and the switch is activated when the voltage value or the current value of the input signal changes. By switching from conduction to non-conduction, the voltage value of the voltage source at that point is stored in the capacitor, and the terminal voltage of the capacitor is output,
This is a non-linear operation circuit that converts an input signal using a conversion function having the same form as the arbitrary function and outputs the converted signal. Further, according to the present invention, a current source whose current value changes as a function of any time including non-linearity is connected to a capacitor via a switch, and the switch is controlled to be conductive and non-conductive by an input signal, The input signal is a pulse phase modulation signal in which a voltage value or a current value changes only at a predetermined time width at a predetermined time point representing input information, and the switch is switched from non-conductive to conductive only within the time width, and By passing a current from a current source to a capacitor in time, the amount of charge stored in the capacitor is increased or decreased by an amount proportional to a value obtained by converting an input signal by a conversion function having the same form as the arbitrary function, and This is a non-linear operation circuit that outputs a terminal voltage.

Further, according to the present invention, in each of the above-mentioned nonlinear arithmetic circuits, a circuit for converting the magnitude of a voltage value or a current value into a pulse modulation signal is provided at a stage preceding the switch, and the input signal is a magnitude of a voltage value or a current value This is a non-linear operation circuit that is an analog signal having information, and is a non-linear operation circuit that temporarily holds the output and feeds it back to the input.

The present invention also provides a voltage source whose voltage value changes as a function of an arbitrary time including non-linearity, a capacitor, and conduction by an input signal disposed between the voltage source and the capacitor and input to an input terminal. The non-linear operation circuit includes a switch whose non-conduction state is controlled, and an output terminal that outputs a voltage of the capacitor. Further, the present invention provides a current source in which a current value changes as a function of arbitrary time including non-linearity, a capacitor, and a conductive or non-conductive state by an input signal arranged between the current source and the capacitor and input to an input terminal. Is a non-linear operation circuit having a switch to be controlled and an output terminal for outputting the voltage of the capacitor. Further, each of the input signals is a non-linear operation circuit which is a pulse width modulation signal or a pulse phase modulation signal. Further, when an analog signal having information on the magnitude of a voltage value or a current value is input, a pulse modulation signal converting means for converting the analog signal into a pulse width modulation signal or a pulse phase modulation signal is connected to each of the input terminals. Is a non-linear operation circuit. A non-linear operation circuit having a means connected to the output terminal for temporarily holding the output voltage of the capacitor, and a feedback loop for feeding back the held capacitor output voltage to the pulse modulation signal converting means instead of the analog signal; It is.

[0031]

DETAILED DESCRIPTION OF THE INVENTION In the present invention, a voltage source that supplies a voltage that varies as a function of time, including non-linearity, is connected to a capacitor via a switch. When the switch is turned on at a timing determined by the input signal, a non-linear operation circuit that converts the input signal by a conversion function having the same form as the arbitrary function and outputs the converted signal is formed. Further, in the above-mentioned nonlinear arithmetic circuit, the output voltage _Vout is fed back to the input via the sample-and-hold circuit, thereby realizing discrete-time nonlinear dynamics of _Vout (t + 1) = F ( _Vout (t)). can do.

Further, in the present invention, by replacing the voltage source with a current source, a non-linear operation for increasing / decreasing a charge amount stored in a capacitor by a very small amount obtained by converting an input signal by a conversion function having the same form as the arbitrary function. Form a circuit.
Also, if the amount of current of the current source is made very small with this circuit,
ΔF = F (x) A small amount of Δx can be generated. Therefore, if the charge of the capacitor is held and the output voltage V _out is fed back to the input through the sample and hold circuit, the nonlinear difference equation ΔF / Δx
= F (x) can be realized. Further, the differential equation dF / dx = F (x) can be approximately solved.

An embodiment of the present invention will be described below with reference to FIGS.
This will be described with reference to FIG. In the following detailed description and the description of the drawings, like elements are denoted by like reference numerals. Embodiment 1 FIGS. 1A and 1B show a first embodiment of the present invention. In the first embodiment, the PWM signal V shown in FIG.
₁ is a non-linear voltage source 3, a switch 4,
And a non-linear conversion circuit including the capacitor 5. Here, the PWM signal input to the input terminal 1 is a pulse width modulation signal having information on the pulse width (T). Simultaneously with the rise of the PWM pulse, the voltage value of the voltage source 3 changes with an arbitrary, for example, a specific non-linear, time function (f (t)) as shown in FIG. Here, it is assumed that the rising time of the PWM pulse is t = 0. The time variation of the output voltage V ₃ of the nonlinear voltage source 3 shown by f (t) in FIG. 1 (d). That is, V ₃ = f (t). While the PWM pulse is 1, the switch 4 is closed, and the voltage from the voltage source 3 charges and discharges the capacitor 5. When the pulse falls (t
= T), the switch 4 opens, and the voltage value (f
(T)) is held in the capacitor and output to the output terminal 2. That is, the output voltage V ₂ = f (T). The operation timing is shown in FIGS. 1B and 1D.

Embodiment 2 In a second embodiment, the PWM input of the first embodiment is replaced with a PPM input, and a PPM signal and a non-linear voltage source are combined.

In the first embodiment, as apparent from the operation principle, the input does not necessarily have to be a PWM signal. It is sufficient that the voltage f (T) can be output at the time T. Therefore, any pulse having a time width enough to charge the capacitor 5 and falling at the time T may be used. This is called a pulse phase modulation (PPM) signal. The operation timing is shown in FIGS. 1 (c) and 1 (d).

Embodiment 3 FIG. 2 shows a third embodiment. In the second embodiment, the input voltage is set to PW
This is a non-linear voltage conversion circuit including a circuit for performing M conversion in a preceding stage. When the input is an analog signal (voltage value V ₁ ) having information on the voltage value, a pulse modulation signal conversion circuit, which is a circuit for converting the voltage of the analog signal into a PWM signal, is used as the input unit of the circuit configuration of the first embodiment. The function of the present invention can be realized by adding it to the preceding stage.

FIG. 2 shows a specific example. First using a comparator 6 the analog input signal V _1, as compared with the lamp voltage V ₁₀ generated by the ramp voltage generating circuit 10, and converts the PWM signal. Here, the lamp voltage refers to a voltage that monotonically increases as shown in FIG. 2B or 2C.
If the change of the lamp voltage is linear as shown in FIG. 2 (b), the pulse width T of the resulting PWM signal is proportional to the input voltage V _1. If the scale is adjusted, from Example 1,
The conversion of V ₂ = f (V ₁ ) can be performed.

As shown in FIG. 2C, if the ramp voltage is made to have a non-linear waveform (V ₁₀ = g (t)), the input voltage V ₁
Is converted to a PWM signal (pulse width T) when T = g ^-1
(V ₁ ) can be performed. When the conversion shown in the first embodiment is subsequently performed, V ₂ = f (T) =
A non-linear conversion product of f (g ⁻¹ (V ₁ )) = f · g ⁻¹ (V ₁ ) can be executed. As a result, even if the functions of f and g are simple, a complicated nonlinear function operation can be performed by a function product. In addition, FIG.
V ₆₁ and (c) is a PWM signal waveform of the output 61 of the comparator 6 in FIG. 2 (a).

FIG. 3A shows an example in which the comparator 6 is configured using the comparator composed of the capacitor and the inverter shown in the conventional example of FIG. FIG. 3B shows the timings of the pulses φ ₁ and φ ₂ , the input voltage V ₁ , and the lamp voltage V
_10, and shows the temporal change of the PWM signal waveform V _61.

Embodiment 4 FIG. 4 shows a fourth embodiment. In the fourth embodiment, after the input voltage is converted into the PWM signal by the ramp generation circuit,
The signal is converted into an M signal, and conversion by a non-linear voltage source is performed using the PPM pulse.

As in the third embodiment, when the input is an analog signal (voltage value V ₁ ) having information on the voltage value, a circuit (6, 6) converts the voltage of the input analog signal into a PPM signal. By adding the pulse modulation signal conversion circuit 10) to the input stage preceding the circuit configuration of the first embodiment, a conversion function using a PPM signal similar to that of the second embodiment can be realized.

For example, in the circuit of Embodiment 3 shown in FIG. 2, as shown in FIG.
The operation based on the PPM signal corresponding to the second embodiment is realized by inserting a pulse source 7, which is a conversion circuit that converts the signal into a signal and outputs the signal, between the comparator 6 and the switch 4. Conversion circuit 7
5 is shown in FIG. The circuit of FIG.
, A direct PWM signal and a PWM signal via inverters 71, 72, 73 connected in series are input. This is a circuit in which the output rises when the input which is the PWM signal falls, and the output falls after the delay time (ΔT) in the inverters 71 to 73.
Therefore, it is a circuit that outputs a pulse of a constant width simultaneously with the fall of the input.

When the PPM signal is generated by the circuit shown in FIG. 5, the falling edge of the PPM signal is lower than the falling edge of the PWM signal shown in FIG. 4B, as shown in FIG. 4C.
Delay by the pulse width (ΔT) of the M signal. If the change of f (t) is sufficiently small within the range of ΔT, there is no problem. If not, the voltage waveform f (t) of the voltage source 10 is also shown in FIG.
It is preferable to start with a delay of ΔT as shown in FIG. Let V ₃ = f (t−ΔT).

Fifth Embodiment In a fifth embodiment, as in the fourth embodiment, the input voltage is subjected to PWM conversion by a ramp generation circuit, and then further to PPM conversion.
The conversion using a non-linear voltage source is performed by using a pulse, and a loop for feeding back an output voltage to an input unit via a buffer is provided. Or, in Example 3 or Example 4, and holds a voltage value by connecting the output voltage V ₂ to the sample-and-hold circuit is it intended to be fed back to the input as the input voltages V ₁ after a certain time.

For example, to make the circuit of the embodiment 3 (a) a feedback type, the output 2 of FIG. 3 (a) may be connected to the input 1 via the buffer 8 as shown in FIG.
This comparator circuit (62, 63, 65, 66, 67, 6)
8) has a sample and hold function, so there is no need to add a new sample and hold circuit. FIG.
(B) shows the timing of the pulses φ ₁ and φ ₂ and the output voltage V
_{2 (V 2 = V 1)} , shows the time change of the lamp voltage V ₁₀ and the PWM signal waveform V _61,. By repeating the sequence shown in FIG. 6B, a discrete-time operation is realized, and the dynamics of V ₂ (t + 1) = f (V ₂ (t)) can be executed.

If a quadratic function is given to V ₃ = f (t) of this circuit, a function equivalent to the circuit shown in FIG. 12 can be obtained.
Obviously, the circuit scale is smaller in this embodiment. Further, this embodiment is superior to the conventional example in that not only a quadratic function but also an arbitrary nonlinear function can be realized.

FIG. 7 shows a branch diagram of a logistic map created by performing a circuit simulation under the same conditions as in FIG. 13 with this circuit. As compared with the result of the conventional circuit shown in FIG. 13, the shape is clearly clear, and is very close to the result of FIG. 11 obtained by precise numerical calculation. This indicates that the configuration of the present circuit is simpler than that of the conventional circuit, so that the circuit is highly accurate.

In this embodiment, the voltage at the terminal 61 is a PWM signal. Assuming that the pulse width is T (t), T (t + 1) = f (T (t)), and the non-linear operation using the PWM signal can be executed simultaneously.

Although the present embodiment is an example in which the circuit of the third embodiment is fed back, a similar result can be obtained by feeding back the circuit of the fourth embodiment. In this case, if the output of the conversion circuit 7 is taken out, a PPM signal is obtained, and a non-linear operation using the PPM signal can be executed simultaneously. Embodiment 6 FIG. 8 shows a sixth embodiment of the present invention. In the circuit according to the fifth embodiment shown in FIG. 6, the voltage of the capacitor 5 is read out via the buffer, and the capacitor 65 constituting the comparator is charged. That you are doing.

Therefore, as shown in FIG. 8A, a circuit is formed by using only one capacitor and using the charge charged by the non-linear operation in the comparator as it is. The operation is as follows.

1) Initially, an appropriate charge is stored in the capacitor 65, or an initial setting is made using a random state such as noise.

2) In response to the clock signal CLK, the nonlinear voltage source 3 and the ramp voltage source 10 each have a voltage V ₃ (t) = f.
(t) and supply of the lamp voltage V ₁₀ (t) are started.
At this time, the switch 62 is closed by the signal S1, and the lamp voltage is supplied to the capacitor 65.

3) The lamp voltage rises and the inverter 66
When the input exceeds the threshold, the inverter is inverted. At this time, the pulse width (time from when CLK is output) appearing at the terminal 61 is defined as T ₁ .

4) As a result, a pulse is emitted from the pulse source 7, the switches 631 and 621 are closed, and the voltage between f (T ₁ ) and the threshold value of the inverter 66 is set across the capacitor 65. . However, pulses from the pulse source 7 for simplicity are sufficiently shorter than the T _1.

5) When the pulse from the pulse source 7 ends, the switch 622 is closed by the signal S2, and the terminal potential of the capacitor 65 is fixed to the power supply voltage VDD, so that the terminal 61 of the inverter remains fixed at the Low level. Becomes

6) When CLK is supplied again, the operation from 2) is repeated.

[0057] Here, the signal S1 is falling down signal rise, the rise of V ₇ by the rise of CLK. Further, S2 is falling down signal rise, the rise of CLK by the fall of the V _7. FIG. 8B shows a clock signal CLK, a non-linear voltage source 3V ₃ (t), a ramp voltage source 10V ₁₀ (t), a signal S1, a signal S2, and a PWM signal V.
₆₁ shows a timing chart of the PPM signal V ₇ from the pulse source 7.

By the above operation, the potential set to the left of the capacitor 65 is f (T _i ) by the pulse width T _i of the terminal 61 at the time i at the discrete time determined by CLK.
Becomes At time i + 1, this potential causes the pulse width T
_{Since i + 1} is determined, by appropriately adjusting the scale, T _{i + 1} = f (T _i ), and the dynamics by an arbitrary function f can be realized.

Since the function of the switch 622 in the above circuit is for fixing the terminal 61 to the low level, a switch may be inserted into the terminal 61 and directly fixed to the low level here.

The circuit shown in FIG. 8A described above does not pass through a buffer which is an analog circuit requiring accuracy, so that the circuit is simplified and the calculation accuracy is improved. That is, this circuit has a great feature that analog dynamics can be realized only by circuit elements that operate digitally. Further, there is an advantage that power consumption can be further reduced because there is no waste of charge and discharge.

Embodiment 7 In Embodiment 7, the operation of Embodiment 6 is performed asynchronously.
That is, in the sixth embodiment, in addition to the synchronous operation by the external clock, the asynchronous operation of starting the next operation by its own output can be realized. That is, a pulse that rises due to the fall of the pulse (V ₇ ) from the pulse source 7 may be set and used as CLK. In this case, the time interval of the pulse exiting from the pulse source 7 is T _i. In addition,
In this circuit, since a nonlinear voltage waveform or a ramp voltage waveform must be started in synchronization with each nonlinear operation circuit, a nonlinear voltage source or a ramp waveform voltage source common to a plurality of nonlinear operation circuits cannot be used.

Embodiment 8 Embodiment 8 is a combination of a PPM input and a non-linear current source instead of using a non-linear voltage source. The eighth embodiment is obtained by replacing the voltage source 3 with a current source in the circuit of the second embodiment shown in FIG. FIG. 9 shows N current sources 31-1.
A circuit having ~ 31-N is shown. This current source 3
1, like the voltage source 3, a current value I ₃₁ is arbitrary, for example of non-linear, but vary in time of the function f (t). That is, it is represented by I ₃₁ = f (t).

Assuming that the pulse width Δt of the PPM signal is sufficiently small and that f (t) hardly changes within the time of Δt, one PPM signal and the capacitor 5 approximately have ΔQ = I ₃₁ Δt = f (T) An amount of charge of Δt is charged (or discharged). Assuming that the capacitance of the capacitor 5 is C,
The change in the output voltage V ₂ is ΔQ / C = f (T) Δt / C, and is proportional to the result of the nonlinear conversion f of the information (time phase T) of the PPM signal.

In the discrete time expression determined by the clock signal serving as the reference of the PPM signal,

(Equation 2)

This means that the dynamics (differential equation with respect to time) is approximately solved. The smaller the pulse width Δt, the more accurate the approximation.

Ninth Embodiment In a ninth embodiment, the input voltage is converted by the non-linear current source after the PPM conversion. In Example 4,
Since the input voltages V ₁ was converted into PPM signals, input to the converter after converting the same input voltage in Example 8 PPM signal. in this case,

(Equation 3)

Is obtained. The product operation of the function shown in the third embodiment can be similarly applied to this embodiment.

Embodiment 10 FIG. 9 shows a tenth embodiment of the present invention. Example 10
Prepares N portions (blocks) excluding the capacitor 5 (N is a positive integer) as shown in the non-linear operation circuits shown in the eighth and ninth embodiments. The input PPM signal of the i-th (i = 1, 2, 3,..., N) block or the voltage input converted to a PPM signal, and the nonlinear current source waveform are represented by V _1-i and f _i (t), respectively. And These blocks are connected to a common capacitor 5. In this circuit, the current I from each block is
_{Since the} capacitor 5 is charged and discharged at _{31-1 (t) to} I _{31-N (t)} ,

(Equation 4)

The above dynamics can be realized. By combining the circuits in this manner, any combination of any non-linear functions can be generated.

In this equation, when it is desired to add or subtract a constant, a constant current source may be used instead of the nonlinear current source (f _i = A, A is a constant). This corresponds to a conventional “switched current source” (reference [2]).

Embodiment 11 Embodiment 11 is for converting an input voltage into a PPM and further using a non-linear current source, and feeding back the output to the input side.

When the ninth embodiment is feedback-connected in the same manner as in the fifth embodiment,

(Equation 5)

The following relationship is obtained. this is

(Equation 6)

This is a circuit that approximately solves the differential equation.

[0075]

In the present invention, only one comparator is used to convert an input voltage into a pulse. The above embodiment has mainly described the voltage input. However, even in the case of a current input, a non-linear conversion circuit can be formed by a circuit exactly the same as that of the above-described embodiment by adding a circuit for storing the current in a predetermined capacitor and converting it to a voltage.

Further, the nonlinear signal source may prepare a waveform having a desired nonlinear / non-monotonic function itself.
There is no need to calculate the inverse function as in the conventional example. Since this non-linear signal source can be used in common for a plurality of circuits, it does not hinder integration. Furthermore, if a non-linear function is used when converting an input signal into a pulse, a product of the non-linear function can be automatically generated. Examples 5 to 7,
As shown in the eleventh embodiment, when the feedback circuit configuration is adopted, not only the voltage output of the operation result but also the PWM
Output can also be provided in the form of signals or PPM signals.

The embodiments of the present invention have been described above. However, the embodiment described here is merely an example, and the embodiment of the present circuit can be variously modified.

[0078]

As described above, according to the present invention, the following effects can be obtained as compared with the conventional circuit shown in FIG.

1) Since the non-linear operation circuit is composed of one comparator, switch and capacitor, the circuit configuration is extremely simple, the area occupied on the chip is small, and the power consumption can be reduced. High integration of the circuit is possible.

2) Since a voltage or current waveform having the same form as a desired nonlinear / non-monotonic function may be given, an arbitrary nonlinear function can be generated and the control circuit can be simplified.

3) Since the circuit is simple, the number of circuit elements causing an operation error is smaller than that of the conventional example, so that the operation accuracy is improved.

4) A current input as well as a voltage input is possible, and furthermore, input / output in the form of a PWM signal or a PPM signal, or simultaneous output thereof, is possible. The configuration can be easily configured.

By utilizing the above characteristics, it can be incorporated in various electric and electronic devices and controllers as a high-performance noise generator and chaos signal generator. Further, it can contribute to advancement of an intelligent processing machine having a recognition function.

[Brief description of the drawings]

FIG. 1 is a diagram showing first and second embodiments of the present invention.

FIG. 2 is a diagram showing a third embodiment of the present invention.

FIG. 3 is a diagram showing a third embodiment of the present invention in which a comparator is shown by a specific circuit.

FIG. 4 is a diagram showing a fourth embodiment of the present invention.

FIG. 5 is a circuit for converting a PWM signal into a PPM signal and outputting the same in the fourth embodiment.

FIG. 6 is a diagram showing a fifth embodiment of the present invention.

FIG. 7 is a branch diagram of a logistic map generated by the circuit of the fifth embodiment.

FIG. 8 is a diagram showing a sixth embodiment of the present invention.

FIG. 9 is a diagram showing a tenth embodiment of the present invention.

FIG. 10 is a diagram showing a conventional nonlinear function generation method.

FIG. 11 is a bifurcation diagram of a logistic map generated by a conventional nonlinear function generation method in a feedback system (numerical simulation result).

FIG. 12 is a diagram showing a feedback circuit configuration using a conventional nonlinear function generation circuit.

FIG. 13 is a bifurcation diagram when a logistic mapping is configured using a conventional nonlinear function generation circuit (circuit simulation (HSPICE) result).

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input terminal 2 ... Output terminal 3 ... Voltage source 4, 62, 63, 621, 622, 631 ... Switch 5, 65 ... Capacitor 6, 601, 602 ... Comparator 7 ... Pulse source 8 ... Buffer 9 ... AND gate 10 ... Lamp voltage generating circuits 31, 39 ... Current sources 66, 67, 68, 71, 72, 73 ... Inverters 74 ... NOR gates

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] August 12, 1999 (1999.8.1)
2)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claims

[Correction method] Change

[Correction contents]

[Claims]

Claims (11)

[Claims] A voltage source whose voltage value changes as a function of an arbitrary time including non-linearity is connected to a capacitor via a switch, and the switch is controlled to be conductive and non-conductive by an input signal. The input signal is a pulse modulation signal having information at the time when the voltage value or the current value changes, and the switch switches from conduction to non-conduction at the time when the voltage value or the current value of the input signal changes, at which time Storing the voltage value of the voltage source in the capacitor and outputting a terminal voltage of the capacitor, thereby converting an input signal by a conversion function having the same form as the arbitrary function and outputting the converted signal. circuit. 2. A current source whose current value changes as a function of an arbitrary time including non-linearity is connected to a capacitor via a switch, and the switch is controlled in conduction and non-conduction by an input signal. The input signal is a pulse phase modulation signal in which a voltage value or a current value changes only at a predetermined time width at a predetermined time point representing input information, and the switch is turned on from a non-conductive state only within a time period of the time width. And by passing a current from the current source to the capacitor within the time, the amount of charge stored in the capacitor in an amount proportional to a value obtained by converting the input signal by a conversion function having the same form as the arbitrary function. Characterized in that the terminal voltage of the capacitor is output. 3. A circuit for converting the magnitude of a voltage value or a current value into a pulse modulation signal is provided before the switch.
2. The non-linear operation circuit according to claim 1, wherein the input signal is an analog signal having information on a voltage value or a current value. 4. A circuit for converting the magnitude of a voltage value or a current value into a pulse phase modulation signal is provided at a stage preceding the switch, and the input signal is an analog signal having information on the magnitude of the voltage value or the current value. 3. The method according to claim 2, wherein
The non-linear arithmetic circuit according to the description. 5. The method according to claim 3, wherein said output is temporarily held and fed back to said input.
The non-linear arithmetic circuit according to the description. 6. The method according to claim 4, wherein said output is temporarily held and fed back to said input.
The non-linear arithmetic circuit according to the description. 7. A voltage source whose voltage value changes as a function of an arbitrary time including non-linearity, a capacitor, and conductive / non-conductive by an input signal disposed between the voltage source and the capacitor and input to an input terminal. A non-linear operation circuit comprising: a switch whose state is controlled; and an output terminal that outputs a voltage of the capacitor. 8. A current source whose current value changes as a function of an arbitrary time including non-linearity, a capacitor, and conduction / non-conduction by an input signal disposed between the current source and the capacitor and input to an input terminal. A non-linear operation circuit comprising: a switch whose state is controlled; and an output terminal that outputs a voltage of the capacitor. 9. The signal according to claim 7, wherein the input signal is a pulse width modulation signal or a pulse phase modulation signal.
9. A nonlinear operation circuit according to claim 8. 10. A pulse modulation signal conversion means for converting an analog signal into a pulse width modulation signal or a pulse phase modulation signal when an analog signal having information on the magnitude of a voltage value or a current value is input. 9. The non-linear operation circuit according to claim 7, wherein the non-linear operation circuit is connected to the non-linear operation circuit. 11. A means for temporarily holding an output voltage of the capacitor connected to the output terminal, wherein the held capacitor output voltage is fed back to the pulse modulation signal converting means instead of the analog signal. 2. The apparatus according to claim 1, further comprising a feedback loop.
0. The non-linear arithmetic circuit according to 0.