SU883927A2 - Device for simulating neurone - Google Patents

Description

The invention relates to the field of bionics and computer technology and can be used as an element of neural networks for modeling biological processes in pattern recognition devices, as well as an element of analyzer structures in robotics.

According to the main author. St. No. 682910, a device for simulating a neuron is known, which contains a synaptic balance changing block, the first inputs of which are the device's inputs, and the outputs are connected to the inputs of the first adder, the second adder and five digital integrators, the output of the first adder is connected to the first input of the first digital integrator, the output of which is connected to the first input of the second adder, the output of which is connected to the first input of the second digital integrator, the output of which is connected by the first inputs of the third and fourth digital integrators, the output of the fourth digital integrator connected to the second input of the second adder, a third input of which is connected to the output of the fifth digital integrator having a first input and a second input

the second integrator is connected to one control input of the device, the second inputs of the n blocks of synaptic balance changes and the second inputs of the first, third, fourth and fifth digital integrators are connected to other control inputs of the device, respectively.

As background information,

When input to the device's inputs when using digital one-bit incrementors in it, pulse streams of one-bit increments, which are 15 pulse sequences of a certain frequency, are used. When digital integrators with multi-bit increments are used in the device, binary D codes are used as initial information entering the device inputs.

However, the device does not allow the processing of visual information coming in the form of light

25 signals directly from the environment, which reduces the functionality of the device and limits the possibilities of using it as an element of visual analysis of mash structures widely used in the design of integrated robots.

The purpose of the invention is to expand the functionality due to the possibility of processing light information.

This goal is achieved by the fact that n receptors have been introduced into the device for modeling of proton, the outputs of which are connected to the first inputs of the n blocks of synaptic balance changes, the inputs of the n receptors are the inputs of the device.

The receptor contains a scale amplifier, a photodiode, a decadal voltage divider, a delay line, a switch, a null organ, a binary counter, and a pulse generator whose output is connected to the counter input of a binary counter, the information outputs of which are connected to the first input of the switch and the output of which is connected to the first input of the null organ, the output of which is connected to the second input of the switch and through the delay line to the installation input of the binary counter, the output of the switch is receptor progress, the cathode of the photodiode is connected to the first input of scaling amplifier whose output is connected to the second input of the zero-body, the anode of the photodiode and the second input of scaling amplifier connected to the zero potential bus.

FIG. 1 is a diagram of the device; FIG. 2 is a diagram of a separate receptor; FIG. 3 - dependence and f {H), in FIG. 4 and 5 time diagrams.

The device for modeling a neuron contains n receptors 1 ,, - lf, the inputs of which are the inputs of the device, n blocks of changing synaptic weights, made in the form of digital integrators, the control inputs connected to the inputs of integrands (second inputs) of digital integrators, the first adder 4, which performs the spatial summation of the input signals, the first digital integrator 5, the input 6 of which is connected to the second input of the integrator 5 and controls the change in the weight of the spatial summation, second Adder 7, the second digital integrator 8, the power of the integral function of which is connected to the input 9, which controls the change in the duration of the time summation. The variable integration input of the integrator 8 is combined with the output of the digital integrator 10 and connected to the input of the variable integration of the digital integrator 11, the output of which is the output of the device, and the input of the integrand function is connected to the input 12, which serves to change the weight value

output value, digital integrator 13, the input of the integrand function of which is connected to 14, the-controlling change of the threshold. The integrators 8 and 10 and the adder 7 form a block of time summation

5 input signals. The inputs of the integration variables of the integrators 10 and 13 are combined and connected to the input 15, which serves to supply increments to an independent variable.

0

The receptor contains a photodiode 16, a scale amplifier 17 with resistors 18 and 19, a null organ 20, a differential amplifier 21 with a resistor 22 in the feedback circuit, a ten-day voltage divider 23, a binary counter 24, a delay line 25, a switch 26 and pulse generator 27,

Consider first the work of a separate receptor (FIG. 2)

0

Photodiode 16 makes it possible to obtain the dependence of the voltage I (obtained at the output of the photodiode) on the illumination H of the external environment. Maximum value.

The photodiode output is very small (on the order of hundreds of millivolts), so for the rest of the circuit to work, it must be amplified. For this purpose, serves the amplifier 17, which has

0 the required gain, which is determined by the value of the resistor 18 included in the negative feedback circuit from the output to the inverted input of the amplifier 17, the voltage from the output of the amplifier 17 goes to the inverted input of the amplifier 21, generator 27, voltage divider 23 and binary counter 24 serve to form a stepped sawtooth

0 voltage The pulses of a rectangular shape of the pulse generator 27 are fed to the input of a binary counter 24, the output of which produces parallel binary numbers equal to the number of pulses calculated

5 counter at a given time. Binary numbers from the output of the counter 24 are fed to the input of the decade voltage divider 23, at the output of which a stepped sawtooth is formed

0 voltage The number of voltage steps in the output signal of voltage divider 23 at a given time is equal to the number of pulses counted by the counter 24 by the same time, i.e. equal to the binary number appearing at the same time at the output of counter 24, the maximum number of steps in the output signal is equal to the recalculation coefficient of counter 24. As soon as the number of counted counters 24 pulses compares with its recalculation coefficient, counter 24 is zeroed and The output of voltage divider 23 is reset to zero. Thus, in the absence of an input signal at the inverted input of the amplifier 21, a stepwise sawtooth voltage is periodically supplied to its direct input from the output of the voltage divider 23.

The output characteristic of the DC decade voltage divider is shown in FIG. four.

A timing diagram explaining the work of the receptor when signals from the external environment enter it is shown in FIG. 5. When light signals come from the external environment, a photodiode 16 at the output of amplifier 17 shows a constant voltage Uy corresponding to the photodiode illumination under the action of these light signals. This voltage is applied .. to the inverted input of the amplifier 21, to the direct input of which a stepped sawtooth voltage is supplied to the output of the voltage divider 23 (Fig. 5a). As soon as the stu: foam sawtooth voltage becomes equal to the voltage value Uy, a voltage drop appears at the output of the amplifier 21, (Fig. 5b). This difference Uy is held until the voltage Uy. will not pass through the delay line 25 (figure 2) and will not reset the contents of the counter 24 to zero. When the counter 24 is zeroed, the value of the output voltage of the voltage divider 23 also drops to zero (Fig. 5a), which entails, in its own, reverse voltage drop at the output of the amplifier 21 (Fig. 56), i.e. . the UNJO voltage also becomes zero.

Thus, a pulse is formed at the output of amplifier 21, the duration of which is determined by the delay time of delay line 25 (Fig. 56) The front and rear edges of this pulse is determined by the value of resistor 22 connected in the positive feedback circuit of amplifier 21 from output to right up my amp. After zeroing the counter 24, the entire process of forming a pulse at the output of the amplifier 21 is repeated. Thus, pulses will appear at the output of amplifier 21 (Fig. 56), the frequency of which is proportional to the illumination of the external environment. The greater the illumination, the greater the pulse frequency at output 21 and vice versa.

The voltage drop appearing at the output 21, with equal voltages at its inputs, goes not only to the input of the delay line 25, but also to the input of the switch 26. (Fig. 2). By this signal, the contents of the counter 24 formed by the time the occurrence of a voltage drop at the output of amplifier 21 is passed by the switch to the receptor output. Only after that the signal from the output of amplifier 21, delayed by the delay line 25, will reset the contents of counter 24 to zero. Thus, at the output of the receptor, one can obtain either a sequence of pulses, the frequency of which is proportional to the illumination of the external environment, or binary numbers, the values of which are also proportional to this illumination.

0

I This is necessary because either a one-bit integrator or a multi-bit integrator can be used in the neuron simulation device.

Information about the illumination of the external environment from the outputs of the receptors Ij (1, ... n) (Fig. 1} goes to the inputs of the integration variables of digital integrators, which are multiplied by the values of the synaptic weights, recorded through the inputs, into the registers of the integrands of these integrators. Obtained products from integrator outputs

S is fed to the inputs of the adder 4, where the spatial summation of the input signals is performed. The resulting amount is fed to the input of integrator 5, which is multiplied by the weight

0 spatial summation, written through the input b in the register of the integrand of this integrator. The product from the output of integrator 5 is fed to the first input of the sum of 5 torus 7, to the second input of which from the output of integrator 13 is fed the product of the values of the threshold and the independent variable, the threshold value is written to the register of the integrator function 13 of the integrator 13, and the increment of the independent variable is fed to the input of the integrator integration variable 13 through the input 15. At the same time point, the result value is temporary

The 5 summation obtained in the previous step (at the first step it is zero) and stored in the register integrator function integrand 10 is multiplied in this integrator by the increment of the independent variable (specified from input 15) and is fed to the input of the integrator integrator variable 8, in which it is additionally multiplied by the duration value

Claims (2)

5 time summation recorded in the integrator integrator 8 register through input 9 and fed to the third input of the adder 7. The time-sum result obtained at this step in the adder is fed to the input of integrator integrand 10, in which the sums are combined with the time sum value obtained in the previous step. After summation, the value of the time sum in the next step is multiplied in integrator 10 by the value of the independent variable coming from input 15. The resulting product from the output of integrator 10 is fed to the input of integrator integration variable 11 which is multiplied by the weight value of the output value written to the integrator integrand register 11 through input 12. If the product obtained in integrator 11 is greater than zero, then a signal appears at the device output. If the product is negative, then there is no signal at the output of the device. Thus, the introduction of a receptor for each of the inputs of the proposed device allows it to perceive and process light, information coming directly from the environment. Claims 1. Device for modeling a neuron according to ed. St. No. 682910, about t and h and it is so that, with the purpose of expanding the functionality due to the possibility of processing light information, n receptors have been introduced in it, the outputs of which are connected to the first inputs n of the synaptic balance changing blocks are device inputs. 2. A device according to claim 1, characterized in that the receptor comprises a large-scale amplifier, a photodiode, a ten-day voltage divider, a delay line, a switch, a null organ, a binary counter and a pulse generator, the output of which is connected to the counting input of the binary counter ,, information outputs which is connected to the first input of the switch and to the inputs of the decade voltage divider, the output of which is connected to the first input of the null organ, the output of which is connected to the second input of the switch and through the delay line to the installation input of the binary the counter output is the output of the switch receptor cathode of the photodiode is connected to the first input of scaling amplifier whose output is connected to the second input of the zero-body, the anode of the photodiode and the second input of scaling amplifier connected to the bus .potentsiala zero. Sources of information taken into account during the examination 1. USSR Author's Certificate No. 682910, cl. G About G 7/60, 1977 (prototype} .. Output organizer PPP PRP P D Lll п п д д