JP5154166B2 - Stochastic resonance ring device, stochastic resonance ring network device, learning memory forgetting device, and attractor selection device - Google Patents

Description

  The present invention relates to a stochastic resonance ring device using stochastic resonance, a stochastic resonance ring network device, a learning memory forgetting device, an attractor selection device, and a stochastic resonance unit.

  In recent years, research for detecting weak signals with high sensitivity using stochastic resonance has been underway (for example, Non-Patent Document 1). Here, stochastic resonance is a phenomenon in which the S / N ratio of an output signal is maximized when a threshold value is processed by adding moderately strong noise to a weak input signal. In the stochastic resonance, in order to maximize the S / N ratio of the output signal, it is necessary to set the noise intensity to an optimum value, but it is not easy to find the optimum value. Therefore, in Non-Patent Document 2, for the purpose of increasing the S / N ratio without optimizing the intensity of the noise signal, a plurality of stochastic resonators 1001 are connected in parallel as shown in FIG. A circuit for adding signals from the element 1001 by an adder 1002 is disclosed.

On the other hand, in recent years, various attempts have been made to realize a signal transmission path in the brain by using stochastic resonance. For example, Non-Patent Document 3 discloses a circuit in which four stochastic resonator units 1004 each having a delay device 1003 added to a stochastic resonator 1001 are unidirectionally coupled in a ring shape as shown in FIG. According to this circuit, since the stochastic resonator 1001 is connected in a ring shape via the delay device 1003, even if the signal input to the circuit is cut off, the signal continues to flow in the circuit for a while. Short-term memory is possible.
Review of Modern Physics, 70,223-287 (1998) (Review of Modern Physics) Nature, 376,236 (1995) (Nature) Phys. Rev. E, 64, 031101 (2001)

  However, the circuit of Non-Patent Document 2 cannot store the signal because the signal disappears at the same time as the signal input is cut off. Further, in the circuit of Non-Patent Document 3, unless the strength of the noise signal in the stochastic resonator 1001 is optimized, the probability that the signal is transmitted by the stochastic resonator 1001 becomes low, and the signal is stored for a long time. Have difficulty.

  An object of the present invention is to provide a stochastic resonance ring device, a stochastic resonance ring network device, a learning memory forgetting device, an attractor selection device, and an attractor selection device capable of storing a signal for a long period of time without optimizing the strength of the noise signal. It is to provide a stochastic resonance unit used in an apparatus.

  The stochastic resonator device of the present invention includes a plurality of stochastic resonators connected in parallel to which signals are respectively input, an adder for adding signals output from the respective stochastic resonators, and a signal added by the adder. And a plurality of stochastic resonance units including a delay unit that outputs the delayed signals, and the plurality of stochastic resonance units are unidirectionally coupled in a ring shape.

  In the stochastic resonator ring device according to the present invention, since a plurality of stochastic resonators are connected in parallel, even if the strength of the noise signal of the stochastic resonator is not optimized, a signal is output by any of the stochastic resonators, It is possible to increase the probability that a signal having a correlation with the signal input to the stochastic resonance unit is transmitted to the stochastic resonance unit connected to the next stage. In addition, since each stochastic resonance unit includes a delay device, even if the signal is temporarily stored and the input signal is interrupted, the signal does not disappear immediately, and the signal having a correlation with the input signal is transmitted for a long time. This signal can be stored for a long time in the device. Furthermore, the probability of signal transmission in the stochastic resonance unit can be adjusted by customizing the number of stochastic resonators connected in parallel and the intensity of the noise signal.

  A stochastic resonance ring network device according to the present invention includes a plurality of stochastic resonance ring devices according to claim 1, wherein the plurality of stochastic resonance ring devices are bidirectionally coupled.

  In the stochastic resonance ring network device according to the present invention, since a plurality of stochastic resonance units are bidirectionally coupled, even if the input signal is interrupted, transmission of a signal correlated with the input signal is continued for a longer period. Can be stored for a longer period of time.

  A learning / memory forgetting device according to the present invention includes the stochastic resonance ring device according to claim 1, and includes m (m is a positive integer) number of first signal transmission units connected in parallel; N (n is a positive integer) second signal transmission units including a stochastic resonance ring network device and connected in parallel to the first signal transmission unit; and the first and second signal transmission units And a stochastic resonance ring device according to claim 1 connected to the output side.

  In the learning and memory forgetting device according to the present invention, if the signal input period becomes long, the probability that the signal is transmitted in all stochastic resonance ring devices increases. A signal is output from the connected stochastic resonance ring device, and the signal is stored. Here, since the stochastic resonance ring network device constituting the second signal transmission unit is composed of a plurality of bi-directionally coupled stochastic resonance ring devices, the stochastic resonance ring device constituting the first signal transmission unit. However, the signal storage period is long. Therefore, in the stochastic ring device on the output side, since the input signal becomes lower as time passes, the output signal gradually attenuates as time passes. Therefore, when a signal is input continuously for a certain period of time, a hardware device that behaves like a human brain, such as learning this signal and gradually forgetting it as time passes. Can be provided.

  An attractor selection device according to the present invention is a signal transmission in which a stochastic resonator, a filter that blocks a direct current component in a signal output from the stochastic resonator, and a stochastic resonator ring device according to claim 1 are connected in series. A plurality of signal transmission units, the plurality of signal transmission units are connected in parallel, and each of the stochastic resonance elements of each signal transmission unit has a different threshold, and the strength of the signal output from one stochastic resonance ring device A suppression coupling unit that couples the respective stochastic resonance ring devices such that the strength of the signal output from the other stochastic resonance ring device is suppressed when the strength of the signal output from the other stochastic resonance ring device is higher It is characterized by providing.

  In the attractor selection device according to the present invention, the threshold value of each stochastic resonator is set to a different value, and the signal output from each stochastic resonator is filtered to remove the DC component and output to each stochastic resonator unit. . Therefore, when a signal having a strength that increases the S / N ratio of a signal output from a certain stochastic resonator is input to the attractor selecting device, the stochastic resonator ring device constituting the signal transmission unit to which the stochastic resonator belongs. A high-level signal is input to the stochastic resonance ring device, a signal is stored for a long period of time, a high-level signal is output from the stochastic resonance ring device, and a signal transmission unit to which the stochastic resonance ring device belongs is selected. It will be. Therefore, by assigning various states to each signal transmission unit in advance, an attractor that selects a suitable state according to the intensity of the input signal, that is, a stable state according to environmental changes. A hardware device that realizes the selection can be provided. Moreover, since the suppression coupling part is provided, this attractor selection device works like a winner-takes-all circuit, and can more reliably select a state of 1.

  Further, the number of the signal transmission units may be three. In this case, three states can be selected according to the intensity of the input signal.

  A stochastic resonance unit according to the present invention includes a plurality of parallel connected stochastic resonators to which signals are respectively input, an adder for adding signals output from the respective stochastic resonators, and a signal added by the adder. And a delay device for outputting after delaying.

  According to this configuration, it is possible to provide a stochastic resonance unit that can store a signal for a long period of time without optimizing the intensity of the noise signal by connecting in a ring shape.

  According to the present invention, a stochastic resonance ring device, a stochastic resonance ring network device, a learning memory forgetting device, and an attractor selection device capable of storing an input signal for a long time without optimizing the intensity of the noise signal, And a stochastic resonance unit used in these devices.

  Hereinafter, a stochastic resonance ring device according to an embodiment of the present invention will be described. FIG. 1 is a diagram showing a stochastic resonance ring device according to this embodiment. As shown in FIG. 1, the stochastic resonance ring device 10 includes four stochastic resonance units 20-1 to 20-4 that are unidirectionally coupled in a ring shape. Note that the stochastic resonance unit 20 is simply referred to when the stochastic resonance unit is not particularly distinguished. In FIG. 1, the number of stochastic resonance units 20 is four, but is not limited thereto, and may be two, three, or five or more.

  Here, the one-way coupling means that the signal flows in one direction. In the case of FIG. 1, the output terminal of the stochastic resonance unit 20-1 is connected to the input terminal of the stochastic resonance unit 20-2, and the input terminal of the stochastic resonance unit 20-1 is connected to the output terminal of the stochastic resonance unit 20-4. The output terminal of a certain stochastic resonance unit 20 is connected to the input terminal of one adjacent stochastic resonance unit 20, and the input terminal of a certain stochastic resonance unit 20 is connected to the other stochastic resonance unit 20 By connecting to the output terminal, each stochastic resonance unit 20 is connected in a ring shape. Therefore, the signal flowing in the stochastic resonance ring device 10 propagates sequentially in the stochastic resonance units 20-1 to 20-4.

  FIG. 2 is a diagram illustrating a configuration of the stochastic resonance unit 20. As shown in FIG. 2, the stochastic resonator unit 20 includes m (m is an integer of 2 or more) stochastic resonators 21-1 to 21-m, an adder 22, and a delay device 23 connected in parallel. . In the case where the stochastic resonator is not particularly distinguished, it is simply referred to as the stochastic resonator 21. FIG. 3 is a diagram showing a configuration of the stochastic resonator 21 shown in FIG. As shown in FIG. 3, the stochastic resonator 21 includes a noise source 211, a mixer 212, and a determination unit 213. The noise source 211 is composed of a function generator that generates various noise signals such as Gaussian white noise and thermal noise. Here, as the intensity of the noise signal, a value that can maintain a certain level of the S / N ratio of the signal output from the stochastic resonator 21 based on the relationship between the approximate level of the input signal and the threshold value. It is preferable to do.

  The mixer 212 superimposes the noise signal output from the noise source 211 and the input signal. The determination unit 213 includes a comparator and an amplifier. The comparator compares the signal output from the mixer 212 with a predetermined threshold value, and outputs a high level signal when this signal is equal to or greater than the threshold value, and outputs a low level signal when this signal is less than the threshold value. To do. In the present embodiment, it is assumed that the threshold value of the comparator is the same in each stochastic resonator 21. The amplifier amplifies the signal output from the comparator with a predetermined amplification factor and outputs the amplified signal to the adder 22. In the present embodiment, the amplification factor of this amplifier is referred to as “coupling constant”. In the present embodiment, the stochastic resonator shown in the upper part of FIG. 3 is represented by a symbol shown in the lower part of FIG.

  Returning to FIG. 2, the adder 22 adds the signals output from the stochastic resonators 21-1 to 21-m and outputs the result to the delay unit 23. The delay unit 23 delays the signal output from the adder for a predetermined time. In the present embodiment, the stochastic resonance unit 20 shown in the upper part of FIG. 2 is represented by a symbol shown in the lower part of FIG.

  In this manner, in the stochastic resonance unit 20 shown in FIG. 2, when a signal is input, the noise signal and the input signal are superimposed by the stochastic resonators 21-1 to 21-m, and the superimposed signal is a threshold value. In the above case, high level signals are output from the respective stochastic resonators 21-1 to 21-m. The signals output from the respective stochastic resonators 21-1 to 21-m are added by the adder 22, then delayed by a delay device 23 for a predetermined time, and output to the adjacent stochastic resonator unit 20.

  Here, in the stochastic resonance unit 20, a plurality of stochastic resonators 21 are connected in parallel, and which stochastic resonator 21 outputs a high-level signal is not determined deterministically but probabilistically. Further, as the level of the input signal increases, the number of stochastic resonators 21 that output a high level signal also increases, so that the stochastic resonator unit 20 outputs a signal having a correlation with the input signal. Is possible. In addition, by increasing or decreasing the number of the stochastic resonators 21, the level of the signal output from the stochastic resonator unit 20 can be adjusted stochastically.

  And since the stochastic resonance unit 20 has a plurality of stochastic resonators 21 connected in parallel, when a signal is input, it can output a high level signal from any of the stochastic resonators 21, A signal having a correlation with the input signal can be transmitted with a high probability without optimizing the intensity of the noise signal of the stochastic resonator 1001 as in the conventional stochastic resonator unit 1004 shown in FIG.

  Next, the operation of the stochastic resonance ring device 10 shown in FIG. 1 will be described. For example, when a signal is input to the stochastic resonance unit 20-1, this signal is probabilistically transmitted by the stochastic resonance units 20-1 to 20-4 and propagates in the circuit. When the input of the signal to the stochastic resonance unit 20-1 is cut off, the signal input to the delay unit 23 immediately before being cut off is delayed by the delay unit 23 and transmitted to the next stage stochastic resonance unit 20. Therefore, the signal does not disappear immediately but propagates in the circuit. On the other hand, since the number of stochastic resonators 21 that output a high level signal decreases as time elapses after the signal input is cut off, the signal circulating in the circuit eventually disappears. In other words, when the signal input is cut off, the signal circulating in the circuit will eventually disappear, but the signal is above a certain level until a certain period has passed since the signal input was cut off. A signal having a correlation with the generated signal circulates in the circuit. Therefore, the stochastic resonance ring device 10 can store the input signal for a long time.

  Next, a stochastic resonance ring network device according to an embodiment of the present invention will be described. FIG. 4 is a diagram showing a configuration of the stochastic resonance ring network device 30. As shown in FIG. As shown in FIG. 4, the stochastic resonance ring network device 30 includes two stochastic resonance ring devices 10-1 and 10-2 that are bidirectionally coupled. Each of the stochastic resonance ring devices 10-1 and 10-2 has the same configuration as the stochastic resonance ring device 10 shown in FIG. 1 and includes four stochastic resonance units 20-1 to 20-4.

  The output terminal of the stochastic resonance unit 20-2 in the stochastic resonance ring apparatus 10-1 is connected to the input terminal of the stochastic resonance unit 20-1 in the stochastic resonance ring apparatus 10-2, and the stochastic resonance of the stochastic resonance ring apparatus 10-2. The output terminal of the unit 20-4 is connected to the input terminal of the stochastic resonance unit 20-3 of the stochastic resonance ring device 10-1. As a result, a signal flowing through the stochastic resonance ring device 10-1 is probabilistically transmitted to the stochastic resonance ring device 10-2, and a signal flowing through the stochastic resonance ring device 10-2 is probable to the stochastic resonance ring device 10-1. The stochastic resonance ring devices 10-1 and 10-2 are bidirectionally coupled.

  Note that the stochastic resonance ring network device 30 illustrated in FIG. 4 includes the two stochastic resonance ring devices 10, but is not limited thereto, and may include three or more stochastic resonance ring devices 10. . Further, the number of stochastic resonance units 20 constituting each stochastic resonance ring device 10 may be different in each stochastic resonance ring device 10.

  FIG. 5 is a diagram illustrating a stochastic resonance ring network device including three stochastic resonance ring devices 10-1 to 10-3. In the case of FIG. 5, the stochastic resonance ring devices 10-1 and 10-2 are bidirectionally coupled, and the stochastic resonance ring devices 10-2 and 10-3 are bidirectionally coupled in cascade. As a result, signals are stochastically transmitted between the stochastic resonators 10-1 and 10-2, and signals are stochastically transmitted between the stochastic resonators 10-2 and 10-3.

  FIG. 6 shows a waveform diagram of a signal that flows through the stochastic resonance ring device 10 when a signal is input to the stochastic resonance ring device 10 shown in FIG. This waveform diagram shows an experimental result by a circuit simulator, (a) shows an input signal, (b) to (g) show a coupling constant ε of the stochastic resonator unit 20 constituting the stochastic resonator ring device 10. , 0.5, 0.39, 0.37, 0.35, 0.30, and 0.1 are shown in FIG.

  In FIG. 6, sin (ωx), which is a sine wave of ω / 2π = 16, is used as an input signal, and this input signal is input during a period of 0 <t <128. A noise signal output from the noise source 211 has a strength of 0.2. Further, the threshold value of the determination unit 213 in each stochastic resonance unit 20 is set to 0.2, and the delay time of the delay unit 23 is set to 64.

  In the period TM1, an input signal is applied to the stochastic resonance ring device 10. Here, a sine wave is adopted as the input signal. As shown in FIG. 6B, when ε = 0.5, no attenuation of the signal circulating in the stochastic resonance ring device 10 was observed during the experiment period. Therefore, it can be seen that the stochastic resonance ring device 10 stores a long-term signal. On the other hand, as shown in FIGS. 6C to 6G, as the coupling constant decreases as ε = 0.39, 0.37.0.35, 0.30, 0.1, the stochastic resonant ring device 10 is used. Although it was observed that the signal circulating inside was rapidly attenuated, it was observed that the required time signal propagated in the stochastic resonance ring device 10 even when the input signal was cut off. Therefore, it can be seen that the signal is stored in the stochastic resonance ring device 10 for a short period of time. From these experimental results, it can be seen that if the coupling constant of the stochastic resonance unit 20 is increased, the signal storage period can be lengthened, and if the coupling constant is decreased, the signal storage period can be shortened.

  FIG. 7 is a waveform diagram of a signal flowing through the stochastic resonance ring network device 30 when a signal is input to the stochastic resonance ring network device 30 shown in FIG. Show. This waveform diagram shows an experimental result by a circuit simulator, (a) shows an input signal, (b) to (g) show a coupling constant ε of the stochastic resonator unit 20 constituting the stochastic resonator ring network device 30. Are waveform diagrams of signals flowing in the circuit when 0.5 is set to 0.5, 0.39, 0.37, 0.35, 0.30, and 0.1, respectively.

  Comparing FIG. 7B to FIG. 7G with FIG. 6B to FIG. 6G, it can be seen that FIG. 7B to FIG. 7B to 7E, no attenuation of the signal circulating in the stochastic ring device 10 is observed during the experiment period, and the signal is stored for a long period of time as in the case of FIG. 6B. You can see that. From these facts, it can be seen that the signal storage period can be lengthened by increasing the number of stochastic resonance ring devices 10 that are bidirectionally coupled.

  FIG. 8 shows a waveform diagram of signals flowing in the device when the number (N) of the stochastic resonator ring devices 10 constituting the stochastic resonator ring network device is changed, and the first stage shows the waveform diagram of the input signal. The second to fifth stages show waveform diagrams in the case of N = 1, 10, 100, and 1000, respectively. The waveform diagram of FIG. 8 is obtained by an experiment using a circuit simulator. In FIG. 8, sin (ωx), which is a sine wave of ω / 2π = 16, is used as an input signal, and this input signal is input during a period of 0 <t <128. A noise signal having a strength of 0.1 is used as a noise signal output from the noise source 211. Further, the coupling constant of each stochastic resonance unit 20 is 0.16, the threshold value of the determination unit 213 is 1, and the delay time of the delay unit 23 is 64.

  As shown in FIG. 8, the signal storage period increases as the number of stochastic resonance units 20 increases from N = 1 to 100. However, in the case of N = 100 and N = 1000, It can be seen that there is no significant difference in the storage period. Therefore, it can be seen that even if the number of stochastic resonance units 20 is 100 or more, an increase in the storage period cannot be expected.

  FIG. 9 shows a waveform diagram of signals flowing through the stochastic resonance ring network device when the number (n) of the stochastic resonance units 20 constituting the stochastic resonance ring network device is changed, and the first stage shows the waveform diagram of the input signal. The second to eighth stages show waveform diagrams in the case of n = 2 to 8, respectively. In FIG. 9, various parameters such as an input signal and a noise signal are the same as those in FIG. Further, the number of stochastic resonator ring devices 10 constituting the stochastic resonator ring network device is 100. As shown in FIG. 9, it can be seen that even if the number of stochastic resonance units 20 is increased, it is not possible to expect a longer storage period.

  Next, a learning / memory forgetting device according to an embodiment of the present invention will be described. FIG. 10 is a diagram showing the configuration of the learning / memory forgetting apparatus according to the present embodiment. This learning memory forgetting device is connected to the output side of two signal transmission units 40-1, one signal transmission unit 40-2, and signal transmission units 40-1 to 40-2 connected in parallel. And a stochastic resonance ring device 10-5.

  The signal transmission unit 40-1 includes a stochastic resonator 21-1, and a stochastic resonator ring device 10-1 connected to the output side of the stochastic resonator 21-1. The stochastic resonant element 21-1 transmits the input signal stochastically, thereby suppressing the signal from being easily output to the stochastic resonant ring device 10-1. Here, the stochastic resonance ring device 10-1 has an output terminal connected to an input terminal of any one of the stochastic resonance units 20 constituting the stochastic resonance ring device 10-5.

  The stochastic resonant ring device 10-1 is the same as the stochastic resonant ring device 10 shown in FIG. 1, and probabilistically transmits the signal output from the stochastic resonant element 21-1 to the stochastic resonant ring device 10-5. Here, the output terminal of any one stochastic resonance unit 20 constituting the stochastic resonance ring device 10-1 is connected to the input terminal of any one stochastic resonance unit 20 constituting the stochastic resonance ring device 10-5. Thus, the stochastic resonance ring device 10-1 and the stochastic resonance ring device 10-5 are connected.

  The signal transmission unit 40-2 includes a stochastic resonator 21-2 and two stochastic resonator devices 10-3 and 10-4 connected to the output side of the stochastic resonator 21-2. The stochastic resonator 21-2 and the stochastic resonator ring device 10-3 are connected in the same manner as the stochastic resonator 21-1 and the stochastic resonator ring device 10-1. The stochastic resonance ring device 10-3 and the stochastic resonance ring device 10-4 are bidirectionally coupled to form a stochastic resonance ring network device 30 shown in FIG. The stochastic resonance ring device 10-4 and the stochastic resonance ring device 10-5 are connected in the same manner as the stochastic resonance ring device 10-1 and the stochastic resonance ring device 10-5.

  The learning memory forgetting device configured in this way operates as follows. When the signal input period is short, the probability that the stochastic resonance ring devices 10-1 to 10-4 transmit the signal is low, and thus the signal storage period of the stochastic resonance ring device 10-5 is shortened.

  On the other hand, when the signal input period becomes long, the probability that the signal is transmitted in all of the stochastic resonance ring devices 10-1 to 10-5 increases. Therefore, a signal is output from the stochastic resonance ring device 10-5 for a long period after the signal input is cut off, and the signal is stored for a long period of time. Here, since the stochastic resonance ring devices 10-3 and 10-4 are bidirectionally coupled, the signal storage period (ignition life) of the stochastic resonance ring devices 10-3 and 10-4 is the stochastic resonance ring device 10. Longer than the signal storage period of -1 and 10-2.

  Therefore, even if the stochastic resonance ring devices 10-1 and 10-2 forget the signal after the signal input is cut off, the stochastic resonance ring devices 10-3 and 10-4 store the signal. A signal is output from the stochastic resonance ring device 10-5. For this reason, the signal output from the stochastic resonance ring device 10-5 gradually decreases.

  FIG. 11 is a waveform diagram showing a signal forgetting process in the learning and memory forgetting device shown in FIG. This waveform diagram shows an experimental result by a circuit simulator. FIG. 11A shows a waveform diagram of the input signal and each of the stochastic resonance ring devices 10-1 to 10-5, the first stage shows the waveform diagram of the input signal, and the second to sixth stages respectively The wave form diagram of the signal output from the stochastic resonance ring apparatus 10-2 to 10-5 is shown.

  11B is an enlarged waveform diagram of the period TM1 shown in FIG. 11A. The first stage shows the waveform of the input signal, and the second stage shows the output from the stochastic resonance ring device 10-5. FIG. 4 shows a waveform diagram of a signal to be generated. In FIG. 11B, the lower waveform diagram has the vertical axis scaled twice as large as the upper waveform diagram.

  FIG. 11C is a waveform diagram in which the period TM2 shown in FIG. 11A is enlarged in the same manner as FIG. 11B. In the period TM21 shown in the upper part of FIG. 11B, a plurality of pulse signals are input as input signals. Thereby, as shown in the second to sixth stages in FIG. 11A, signals are stored by the stochastic resonance ring devices 10-1 to 10-5. As shown in the second to sixth stage waveform diagrams of FIG. 11A, the stochastic resonator ring devices 10-3 and 10-4 are bi-directionally coupled to form a stochastic resonator ring network device. It can be seen that the signal decay time is longer and the storage period is longer than the resonant ring devices 10-1, 10-2, 10-5.

  In the cutoff period TM22 in which the input signal is cut off, as shown in the lower part of FIG. 11B, the level of the signal output from the stochastic resonance ring device 10-5 is attenuated as time elapses. I understand that. When a pulse signal of one pulse is input as an input signal at time T1 when the cutoff period TM22 has elapsed, a high-level signal is output from the stochastic resonance ring device 10-5 so as to respond to the pulse signal at time T2. It can be seen that is output from. This is because the stochastic resonance ring devices 10-1 to 10-5 store this pulse signal by inputting a plurality of pulse signals in the period TM21, and another pulse signal is stored in the state where the pulse signal is stored. This is because it was entered.

  On the other hand, as shown in the upper part of FIG. 11C, in the period TM31, a plurality of pulse signals are input as input signals in the same manner as in the period TM21, and at a time T3 when a cutoff period TM32 longer than the cutoff period TM22 has elapsed. One pulse signal is input. Then, as shown in the lower part of FIG. 11 (c), at time T4, a signal is output from the stochastic resonance ring device 10-5 so as to respond to this pulse signal. The level of this signal is shown in FIG. 11 (b). It is considerably lower than the case. This is because the cutoff period T32 is longer than the cutoff period T22, and another pulse signal is input in a state where the stochastic resonance ring devices 10-1 to 10-5 forget the signal.

  As described above, according to the learning and memory forgetting device, when a pulse signal for a plurality of pulses is input continuously for a certain period of time, the pulse signal is learned and the pulse signal is gradually forgotten over time. Thus, it is possible to provide a hardware device that behaves like a human brain.

  In the learning memory forgetting device shown in FIG. 10, the number of signal transmission units 40-1 is two and the number of signal transmission units 40-2 is one. However, the present invention is not limited to this. -1 may be one or three or more, and the number of signal transmission units 40-2 may be two or more.

  In addition, the signal transmission unit 40-2 is configured by a stochastic resonant ring network device including two stochastic resonant ring devices 10 that are bi-directionally coupled. A stochastic resonant ring network device including the resonant ring device 10 may be used.

  Next, an attractor selection device according to an embodiment of the present invention will be described. FIG. 12 is a diagram showing a configuration of the attractor selection device according to the embodiment of the present invention. The attractor selection device includes three signal transmission units 50-1 to 50-3. Note that the signal transmission units 50-1 to 5-3 are simply referred to as the signal transmission unit 50 when not particularly distinguished. Further, the number of signal transmission units 50 is not limited to three, and may be two or four or more.

  The signal transmission unit 50-1 includes a stochastic resonance element 21-1, a filter 60-1, a stochastic resonance ring device 10-1, and an output terminal TT1-1. The signal transmission unit 50-2 includes a stochastic resonance element 21-. 2, a filter 60-2, a stochastic resonance ring device 10-2, and an output terminal TT 1-2, and a signal transmission unit 50-3 includes a stochastic resonance element 21-3, a filter 60-3, a stochastic resonance ring device 10- 3 and output terminals TT1-3.

  Note that when the filters 60-1 to 60-3 and the output terminals TT1-1 to TT1-3 are not particularly distinguished, the filter 60 is simply referred to as the output terminal TT1. The stochastic resonance element 21-1 transmits the input signal stochastically, thereby suppressing the signal from being easily transmitted to the stochastic resonance ring device 10-1. The filter 60-1 cuts the DC component of the signal output from the stochastic resonator 21-1, thereby allowing the component having a high S / N ratio to pass therethrough and outputting it to the stochastic resonator device 10-1. The filter 60-1 is the same as the filters 60-2 and 60-3, and the stochastic resonance ring device 10-1 is the same as the stochastic resonance ring devices 10-2 and 10-3. The output terminals TT1-1 to TT1-3 are connected to the stochastic resonance ring devices 10-1 to 10-3, respectively, and output signals flowing through the stochastic resonance ring devices 10-1 to 10-3.

  The threshold values of the stochastic resonators 21-1 to 21-3 are set higher in this order. That is, the threshold value of the stochastic resonator 21-1 is minimum, the threshold of the stochastic resonator 21-3 is maximum, and the threshold of the stochastic resonator 21-2 is intermediate between the thresholds of the stochastic resonators 21-1 and 21-3. Is set to the value of

  Therefore, the relationship between the S / N ratio and the intensity of the signals output from the stochastic resonators 21-1 to 21-3 is as shown in FIG. 13 when the intensity of the noise signal is constant. FIG. 13 is a graph showing the relationship between the S / N ratio and intensity of signals output from the stochastic resonators 21-1 to 21-3. In this graph, curves C1 to C3 correspond to the stochastic resonators 21-1 to 21-3, respectively.

  Since the stochastic resonator 21-1 has the smallest threshold, it can be seen that the S / N ratio is high with respect to a weak signal as shown by the curve C1. In addition, since the stochastic resonator 21-3 has the largest threshold, it can be seen that the S / N ratio is high with respect to a strong signal as indicated by the curve C3. Further, since the threshold value of the stochastic resonator 21-2 is an intermediate value, it can be seen that the S / N ratio is high with respect to a signal having an intermediate intensity, as shown by the curve C2.

  Therefore, the S / N ratio of the signal output from the stochastic resonator 21-1 as an input signal is higher than the S / N ratio of the signals output from the other stochastic resonators 21-2 and 21-3. When a weak signal is input, if the high level and low level are encoded with “1” and “0”, the signals of “1” and “0” are repeatedly output from the stochastic resonator 21-1. Then, signals of “0” are mainly output from the other stochastic resonators 21-2 and 21-3. Therefore, the intensity of the signal input to the stochastic resonance ring device 10-1 is the highest among the signals input to the stochastic resonance ring devices 10-1 to 10-3.

  In addition, a signal having an intermediate intensity is input as an input signal so that the S / N ratio of the signal output from the stochastic resonator 21-2 is higher than those of the other stochastic resonators 21-1, 21-3. Then, the stochastic resonator 21-2 repeatedly outputs “1” and “0” signals, and the stochastic resonator 21-1 mainly outputs “1” signals from the stochastic resonator 21-3. Mainly outputs a signal of “0”. Therefore, the DC component of the signal output from the stochastic resonator 21-1 is cut by the filter 60-1, and among the signals input to the stochastic resonator devices 10-1 to 10-3, the stochastic resonator device 10- The intensity of the signal input to 2 is the highest.

  Further, when a signal having a strong intensity is inputted as an input signal, the S / N ratio of the signal output from the stochastic resonator 21-3 is higher than those of the other stochastic resonators 21-1, 21-2. The stochastic resonators 21-3 repeatedly output “1” and “0” signals, and the stochastic resonators 21-1 and 21-2 mainly output “1” signals. Therefore, the DC components of the signals output from the stochastic resonators 21-1 and 21-2 are cut by the filters 60-1 and 60-2, and the signals input to the stochastic resonators 10-1 to 10-3 Among them, the intensity of the signal input to the stochastic resonance ring device 10-3 is the highest.

  Returning to FIG. 12, each of the stochastic resonance ring devices 10-1 to 10-3 includes suppression coupling units 101-1 to 101-3. In addition, when not particularly distinguishing the suppression coupling units 101-1 to 101-3, the suppression coupling unit 101 is simply used. The suppression coupling unit 101-1 includes output terminals TT1-2 and T1-3 of signal transmission units 50-2 and 50-3 different from the signal transmission unit 50-1 to which the suppression coupling unit 101-1 belongs, and the stochastic resonance ring device 10-1. It is connected between the stochastic resonance unit 20-3. Note that the connection relationship between the suppression coupling units 101-2 and 101-3 is the same as the connection relationship between the suppression coupling unit 101-1, and thus the description thereof is omitted.

  FIG. 14 is a diagram illustrating a configuration of the suppression coupling unit 101-1. The suppression coupling unit 101-1 includes an amplifier, the input side is connected to the output terminals TT1-2 and TT1-3, and the output side is each determination of the stochastic resonators 21-1 to 21-m of the stochastic resonator unit 20-3. Connected to the unit 213.

  Here, the determination unit 213 includes a comparator that increases the threshold value as the level of the signal output from the suppression coupling unit 101-1 increases. Therefore, the suppression coupling unit 101-1 can adjust the threshold value of each of the stochastic resonators 21-1 to 21-m constituting the stochastic resonator unit 20-3 according to the level of the input signal. The suppression coupling units 101-2 and 101-3 have the same configuration as that of the suppression coupling unit 101-1, and thus the description thereof is omitted. Further, the suppression coupling unit 101-1 is not limited to the stochastic resonance unit 20-3, and may be connected to at least one of the stochastic resonance units 20-1 to 20-4.

  The attractor selection device configured as described above operates as follows. In the following description, it is assumed that a signal having an intermediate intensity that maximizes the S / N ratio of the stochastic resonator 21-2 is input as the input signal. When an intermediate strength signal is input, a signal “1” is mainly output from the stochastic resonator 21-1, and signals “1” and “0” are repeatedly output from the stochastic resonator 21-2. Then, a signal of “0” is output from the stochastic resonator 21-3. Since the DC components of these signals are cut by the filters 60-1 to 60-3, the intensity of the signal input to the stochastic resonance ring device 10-2 is the highest. Therefore, the stochastic resonance ring device 10-2 stores a signal for a longer time than the other stochastic resonance ring devices 10-1 and 10-3, and the level of the signal output from the output terminal TT1-2 is the highest.

  Furthermore, the threshold value of the stochastic resonance unit 20-3 of the stochastic resonance ring device 10-2 is changed to the stochastic resonance unit of the other stochastic resonance ring devices 10-1 and 10-3 by the action of the suppression coupling units 101-1 to 101-3. The signal output from the stochastic resonator device 10-2 is increased, the signals output from the stochastic resonator devices 10-1 and 10-3 are suppressed, and the output terminal TT1- Only 2 will be selected.

  That is, this attractor selection device works like a winner-takes-all circuit in which only the output terminal TT1-2 that outputs the maximum signal is selected and the other output terminals TT1-1 and TT1-3 are not selected.

  Therefore, when the input signal changes from a weak signal to an intermediate signal, or when the input signal changes from an intermediate signal to a small signal, either of the output terminals TT1-1 and T1-2. The attractor selection device will have hysteresis such that the thresholds that determine whether is selected are different.

  Even when a weak or strong signal that increases the S / N ratio of the stochastic resonator 21-1 or 21-3 is input as an input signal, the output terminal TT1- The level of the signal output from 1 or TT1-3 is increased, and the output terminal TT1-1 or TT1-3 is selected.

  Thereby, by assigning certain states A to C to the signal transmission units 50-1 to 50-3, a stable state is selected according to the intensity of the input signal, that is, a stable state according to a change in environment. It is possible to provide a hardware device that realizes attractor selection for selecting the.

  The present invention is useful for realizing a coprocessor that performs flexible processing or ambiguous processing in the field of next-generation control systems and information processing systems that control robots in a fluctuating environment.

  In addition, since the present invention can be driven by thermal noise at room temperature, it is possible to reduce the energy required for the threshold judgment process by selecting an appropriate material, which is extremely low power consumption that could not exist at all. It is possible to realize a brain-type information processing apparatus that operates with electric power.

It is the figure which showed the stochastic resonance ring apparatus by this Embodiment. It is a figure which shows the structure of the stochastic resonance unit shown in FIG. It is a figure which shows the structure of the stochastic resonator shown in FIG. It is the figure which showed the structure of the stochastic resonant ring network apparatus by which the stochastic resonant ring apparatus shown in FIG. 1 was bi-directionally coupled. It is the figure which showed the stochastic resonance ring network apparatus comprised from three stochastic resonance ring apparatuses. FIG. 2 is a waveform diagram of a signal that flows through the stochastic resonance ring device when the signal is interrupted after a signal is input to the stochastic resonance ring device shown in FIG. 1 for a certain period of time. FIG. 6 shows a waveform diagram of a signal that flows through the stochastic resonance ring device when a signal is input to the stochastic resonance ring device shown in FIG. The wave form diagram of the signal which flows through a stochastic resonance ring apparatus when changing the number of the stochastic resonance ring apparatuses bi-directionally coupled is shown. FIG. 5 shows a waveform diagram of a signal flowing through the stochastic resonance ring device when the number of stochastic resonance units constituting the stochastic resonance ring device is changed. It is the figure which showed the structure of the learning memory forgetting device by this Embodiment. It is a wave form diagram which shows the forgetting process of the signal in the learning memory forgetting device shown in FIG. It is a figure which shows the structure of the attractor selection apparatus by embodiment of this invention. It is the graph which showed the relationship between S / N ratio of the signal output from each stochastic resonator, and intensity | strength. It is a figure which shows the structure of a suppression coupling | bond part. It is the figure which showed the conventional circuit. It is the figure which showed the conventional circuit.

Explanation of symbols

10 10-1 to 10-4 Stochastic resonance ring device 20 20-1 to 20-4 Stochastic resonance unit 21 21-1 to 21-m Stochastic resonance element 22 Adder 23 Delay device 30 Stochastic resonance ring network device 40 40-1 ˜40-2 Signal transmission unit 50 50-1 50-3 Signal transmission unit 60 60-1 to 60-3 Filter 101 Suppression coupling unit 211 Noise source 212 Mixer 213 Determination unit

Claims (5)

A plurality of stochastic resonators connected in parallel to which signals are respectively input, an adder for adding signals output from the respective stochastic resonators, and a delay device for delaying and outputting the signals added by the adders Including a plurality of stochastic resonance units including
The stochastic resonance ring device, wherein the plurality of stochastic resonance units are unidirectionally coupled in a ring shape.   A stochastic resonance ring network device comprising a plurality of stochastic resonance ring devices according to claim 1, wherein the plurality of stochastic resonance ring devices are bidirectionally coupled. M (where m is a positive integer) first signal transmission units including the stochastic resonance ring device according to claim 1 and connected in parallel;
An n (n is a positive integer) number of second signal transmission units connected in parallel to the first signal transmission unit, including the stochastic resonance ring network device according to claim 2;
A learning memory forgetting device comprising: the stochastic resonance ring device according to claim 1 connected to the output side of the first and second signal transmission units. A plurality of signal transmission units each including a stochastic resonator, a filter that blocks a direct current component in a signal output from the stochastic resonator, and the stochastic resonator ring device according to claim 1 are connected in series.
The plurality of signal transmission units are connected in parallel,
A different threshold is set for each of the stochastic resonators of each signal transmission unit,
When the strength of a signal output from one stochastic resonator ring device is higher than the strength of a signal output from another stochastic resonator ring device, the strength of a signal output from the other stochastic resonator ring device is suppressed. An attractor selection device further comprising a suppression coupling unit coupling the stochastic resonance ring devices. The attractor selection device according to claim 4, wherein the number of the signal transmission units is three.