Bionic auditory localization system based on bridge artificial neuron units of memristor circuit

Abstract

The embodiment of the application discloses a kind of bridge artificial neuron unit composition's bionic hearing positioning system based on memristor circuit, it is related to neural morphic computing and artificial intelligence hardware field.In the present application: bridge artificial neuron unit includes the first memristor analog circuit connected with first capacitor and first load resistance, and the second memristor analog circuit connected with second capacitor and second load resistance, which are connected by first bridge capacitor, and coupled with external circuit by second bridge capacitor.The unit can simulate the pulse emission and refractory behavior of biological neuron.A plurality of the unit can be connected in series to build a one-way pulse propagation chain, forming a pulse propagation network.Based on the network, a coincidence detector can detect the small time difference of multiple input signals and be applied to a bionic hearing system to achieve sound source positioning.The present application has simple structure, high degree of modularity, and can realize high-reliability analog pulse propagation and high-precision coincidence detection.

Description

Technical Field

[0001] This invention relates to the field of neuromorphic computing and artificial intelligence hardware, and in particular to a biomimetic auditory localization system based on memristor circuits and composed of bridging artificial neuron units. Background Technology

[0002] Biological nervous systems achieve efficient information processing through precise temporal integration. Among these, the pulse propagation and coincidence detection of action potentials play a crucial role in neural coding and signal processing. Coincidence detection, a mechanism for integrating multiple neural signals arriving within a short period, is particularly important in fields such as auditory localization. In recent years, artificial neurons based on devices such as memristors have successfully simulated the "all-or-none" pulse firing behavior of biological neurons. Current research on artificial neurons mainly focuses on mimicking the spike behavior described by typical models such as the Hodgkin-Huxley model and the leakage integral discharge model. Silicon-based simple leakage integral discharge model neurons require a large number of components and complex circuits to simulate the complex ion channel dynamics. In contrast, threshold-switched memristors, due to their negative differential resistance and dynamic impedance characteristics, are considered ideal electronic simulators for simulating the ion channel behavior of biological neurons. However, most existing artificial neurons can only achieve single-point firing, making it difficult to construct neural networks capable of reliably propagating temporal signals on a hardware architecture. Therefore, it is difficult to accurately replicate the delayed propagation characteristics of action potentials on neural axons in biological neurons, thus limiting their development in applications requiring spatiotemporal information processing.

[0003] The time sensitivity of neural signals is crucial for information processing in neuronal systems. Coincidence detection, a mechanism that integrates neural activity from different neurons within a short time window, responds significantly stronger to coincident input spikes than to non-coincident input spikes. It is widely involved in neural coding and perception, and is closely related to auditory processing. For example, coincidence detection neurons in the auditory brainstem of mammals and birds utilize interstitial delay (ITD) for sound localization. Mammals primarily rely on the medial superior olive (MSO) for ITD processing, while birds perform this function in the laminar nucleus (NL). The Jeffress model explains ITD-based sound localization through a mechanism involving delay lines and coincidence detection. According to this model, neurons in the auditory system receive sound signals from both ears through delay lines of different lengths, creating a small time difference. When a neuron receives precisely delayed inputs from both ears, it generates an action potential, enabling coincidence detection to determine the direction of the sound source. Although artificial spike neurons have been successfully developed, the simulation of spike propagation, crucial to the Jeffress model, remains largely unexplored. The main challenge lies in the lack of modular spike units capable of reliably propagating signals within the network.

[0004] A few studies have also investigated highly compact transistor-based neural circuits to simulate the spike behavior of biological neurons. By employing modular structures, spike neural networks capable of coincidence detection based on the Jeffress model can be constructed. However, the inter-module connections used for spike propagation are primarily resistive, and these methods heavily rely on complex circuit designs and a large number of components.

[0005] In summary, existing solutions typically rely on complex circuit designs and bulky hardware, resulting in low scalability and integration. Therefore, there is a need for a simple and highly modular artificial neuron unit and its associated overlap detection system to achieve highly reliable analog pulse propagation and high-time-accuracy overlap detection, thereby realizing a biomimetic neural auditory localization system. Summary of the Invention

[0006] The embodiments of the present invention provide a biomimetic auditory localization system based on a memristor circuit and a bridged artificial neuron unit, which can achieve highly reliable analog pulse propagation and high temporal accuracy of coincidence detection.

[0007] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0008] In a first aspect, embodiments of the present invention provide a bridged artificial neuron unit based on a memristor circuit, comprising: simulating ion channels of a biological neuron through two branches, each branch including a memristor simulation circuit, a capacitor, and a load resistor; connecting a bridging capacitor between the two branches; and the bridged artificial neuron unit being connected to an external circuit through another bridging capacitor.

[0009] The first memristor analog circuit M1, which has threshold switching characteristics, is connected to a first capacitor Cm1 and a first load resistor Rm1 at both ends, forming the first branch of the simulated sodium ion channel; wherein, both the first memristor analog circuit M1 and the second memristor analog circuit M2 have threshold switching characteristics, and the bridging artificial neuron unit can generate pulse current signals under the excitation of an external input signal.

[0010] The second memristor analog circuit M2, which has threshold switching characteristics, is connected to a second capacitor Cm2 and a second load resistor Rm2 at both ends, forming a second branch simulating a potassium ion channel; a first bridging capacitor Cb1 is connected between the first branch and the second branch; and a second bridging capacitor Cb2 is connected between the first branch and / or the second branch and an external circuit.

[0011] Specifically, the first memristor analog circuit M1 and the second memristor analog circuit M2 have the same structure, including: a PNP type transistor and an NPN type transistor with their bases and collectors connected in series; a base resistor Rb connected between the bases of the PNP type transistor and the NPN type transistor; a parallel resistor Rp connected between the emitters of the PNP type transistor and the NPN type transistor; and a feedback resistor Rs connected to the emitter terminal of the NPN type transistor. A constant voltage source excites the bridged artificial neuron unit based on the memristor circuit to generate periodic current pulses.

[0012] Secondly, embodiments of the present invention provide a biomimetic auditory localization system based on a memristor circuit-bridged artificial neuron unit, comprising:

[0013] A pair of sound signal acquisition modules are used to simulate sound signals from the left and right ears and convert them into electrical signals of a pulse sequence. The pulse sequence is received and propagated by a pulse propagation network, and a specific delay is introduced to compensate for a specific binaural time difference. The input of the coincidence detector is connected to the output of the corresponding pulse propagation network. When the two delayed pulse signals coincide in time, the output unit of the corresponding coincidence detector generates an output pulse to achieve sound source localization.

[0014] The pulse propagation network is composed of at least two bridging artificial neuron units connected in series; wherein the output terminal of the second bridging capacitor of the preceding bridging artificial neuron unit is connected to the input terminal of the following bridging artificial neuron unit, so that the pulse signal propagates unidirectionally along the series connection direction of the at least two bridging artificial neuron units. Optionally, the structure type of the pulse propagation network includes a chain structure, a tree-like branching structure, and a two-dimensional grid structure.

[0015] Specifically, the coincidence detector includes one output unit and two input units, with the output of the input unit bridged to the input of the output unit; the specific structure of the output unit and the input unit is as described above in the bridging artificial neuron unit; the input of the coincidence detector is bridged to the output of the pulse propagation network to receive the delayed multiple signals; when the multiple signals coincide in time, the output channel of the coincidence detector generates an output pulse.

[0016] The weights and time integration windows of each input signal are adjusted by adjusting the load resistor Rm1 parameter in the output unit and / or the second bridging capacitor Cb2 parameter in the input unit. The bionic auditory localization system based on a memristor circuit-based bridging artificial neuron unit provided in this embodiment includes a first memristor analog circuit connected to a first capacitor and a first load resistor, and a second memristor analog circuit connected to a second capacitor and a second load resistor. The two are connected through the first bridging capacitor and coupled to an external circuit through the second bridging capacitor. This unit can simulate the pulse firing and refractory period behavior of biological neurons. Multiple units connected in series can construct a unidirectional pulse propagation chain, forming a pulse propagation network. An overlap detector constructed based on this network can detect minute time differences between multiple input signals and is applied to a bionic auditory system to achieve sound source localization. This realizes a simple and highly modular artificial neuron unit and its associated overlap detection system, achieving highly reliable simulated pulse propagation and high-time-accuracy overlap detection. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram illustrating the structure and characteristics of the bridging artificial neuron unit of the present invention, wherein: Figure 1 a shows the structure of biological neurons and ion channels; Figure 1 b shows a simplified schematic diagram and specific circuit structure of the bridging artificial neuron unit of the present invention; Figure 1 c shows the specific structure of the memristor analog circuit; Figure 1 d shows the current-voltage and voltage-current characteristic curves of this memristor analog circuit; Figure 1 e shows an RC oscillator circuit composed of memristor analog circuitry; Figure 1 f shows the pulse current sequence generated by the oscillator under a constant voltage input.

[0019] Figure 2 This is a schematic diagram and experimental results illustrating the simulation of the refractory period of biological neurons and the construction of a pulse propagation network according to the present invention, wherein: Figure 2 a shows the propagation process of action potentials in biological neurons; Figure 2 b shows the pulse propagation chain composed of cascaded bridging artificial neuron units; Figure 2 c shows the specific circuitry of the stimulation module and the artificial neuron unit in the chain; Figure 2 d shows the irregular phenomenon of the output frequency changing with the input frequency; Figure 2 e shows a comparison of the input and output pulse waveforms under a specific stimulus voltage; Figure 2 f is a simplified schematic diagram of the pulse propagation process; Figure 2 g shows the pulse signals synchronously acquired at different locations in the propagation chain; Figure 2 h shows the pulse propagation network structure composed of horizontal and vertical connections between units; Figure 2 i shows the pulse signal propagation process at different locations in the network.

[0020] Figure 3 The diagram shows the overlap detector constructed based on bridging artificial neuron units and its detection performance, wherein: Figure 3 a demonstrates the principle of detecting neurons through biological overlap; Figure 3 b shows the "dual-input-single-output" coincidence detector structure constructed from the units of the present invention; Figure 3 c shows the specific circuitry of the synaptic structure in the detector; Figure 3 d is a simplified diagram of the overlap detection process; Figure 3 e shows the relationship between the output pulse frequency and the input phase difference; Figure 3 f shows the input and output pulse waveforms under a specific phase difference.

[0021] Figure 4 This is a diagram illustrating the structure and function of the bionic auditory positioning system of the present invention, wherein: Figure 4 a demonstrates the biological principle by which barn owls use the time difference between their two ears to locate sound sources; Figure 4 b shows a block diagram of the bionic hearing system of the present invention, which uses delay lines and coincidence detectors to process ITD (Interaural Time Difference) information; Figure 4 c shows the system's recognition results for different ITDs (corresponding to different sound source directions). Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Embodiments of the present invention will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of the present invention means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0023] This invention provides a bridging artificial neuron unit, a pulse propagation network, an overlap detection system, and a bionic auditory application based on a memristor circuit. This solution has the advantages of simple structure, high biosimilarity, and strong scalability. To achieve the above objectives, a bridging artificial neuron unit was designed and developed, and based on this unit, a pulse propagation network, an overlap detector, and a finalized bionic auditory localization system were further designed.

[0024] The bridging artificial neuron unit is based on two memristor circuits that simulate the sodium and potassium ion channels of a biological neuron, respectively, and are coupled by a bridging capacitor. Under external voltage stimulation, this unit can generate current pulses similar to the action potential of a biological neuron and exhibits refractory period characteristics.

[0025] The pulse propagation network is composed of multiple bridging artificial neuron units connected in series. It can achieve attenuation-free, unidirectional propagation of pulse signals with a propagation delay on the order of microseconds, simulating signal conduction on biological axons.

[0026] The coincidence detector uses the aforementioned pulse propagation network as a delay line to delay at least two input signals. When these signals coincide in time (i.e., the time difference is less than a certain threshold), it drives an output neuron unit to generate a pulse response. The frequency of the output pulse is closely related to the phase difference between the input signals, demonstrating accurate time detection capability.

[0027] The biomimetic auditory localization system utilizes the aforementioned coincidence detector to simulate the Jeffress model in biological auditory systems. By setting up delay line arrays with different delay values, the system can convert the time difference between the two ears into neuronal activations at different spatial locations, thereby achieving sound source localization and exhibiting good noise immunity.

[0028] The following are specific examples for illustration:

[0029] Example 1: Bridging Artificial Neuron Unit

[0030] like Figure 1 As shown, the bridging artificial neuron unit includes a first memristor analog circuit M1, a second memristor analog circuit M2, a first capacitor Cm1, a second capacitor Cm2, a first bridging capacitor Cb1, a second bridging capacitor Cb2, a first load resistor Rm1, and a second load resistor Rm2.

[0031] M1 is connected in parallel with Cm1 (0.1μF) and in series with Rm1 (2200Ω) to form a simulated branch of the sodium ion channel.

[0032] M2 is connected in parallel with Cm2 (0.1μF) and in series with Rm2 (2200Ω) to form a simulated branch of the potassium ion channel.

[0033] The two branches are connected by a bridging capacitor Cb1 (0.1μF).

[0034] The unit is coupled to the external circuit through the second bridging capacitor Cb2 (0.1μF).

[0035] The specific structures of memristor analog circuits M1 and M2 are as follows: Figure 1 As shown in section c, it consists of a PNP transistor (MMBT3906), an NPN transistor (MMBT3904), and three types of resistors Rb (1kΩ), Rp (200kΩ), and Rs (200Ω), exhibiting S-type negative differential resistance characteristics and threshold switching characteristics. Figure 1 (part d).

[0036] Under constant voltage source excitation (e.g., 1.3V), this unit can generate a periodic current pulse sequence. Figure 1 (f part), simulating the pulse firing of neurons.

[0037] Example 2: Pulse Propagation Network

[0038] like Figure 2 As shown, refer to Figure 2 In section bc, multiple neuronal units from Example 1 are connected in series to form a pulse propagation chain. The left side is the stimulation module (structure similar to the potassium ion channel branch in the unit), and the right side is the artificial axon (composed of multiple units connected in series).

[0039] Experiments show that ( Figure 2 (de part) When the input pulse frequency is low (<1000 Hz), the output frequency is basically the same as the input frequency. When the input frequency is too high, the output frequency saturates at about 850 Hz, exhibiting a refractory period phenomenon similar to that of biological neurons. Figure 2 The g-part shows that the waveform and amplitude of the pulse remain stable as it propagates through the chain, with no significant attenuation, and each unit introduces a propagation delay of about 80 μs, which is comparable to the delay of biological axon nodes.

[0040] Example 3: Overlap Detector

[0041] like Figure 3 As shown, refer to Figure 3 In the bc section, a dual-input single-output coincidence detector is constructed. The two upper chains serve as inputs (to simulate presynaptic neurons), and the lower chain serves as the output (to simulate postsynaptic neurons). The signal weights and integration time window are adjusted by modifying the capacitance (Cb2'=0.02μF) in the input branch and the resistance (Rm1'=2550Ω) in the output branch.

[0042] A sinusoidal voltage signal (1.6±0.2V) with a frequency of 500 Hz but a phase difference ΔPhase is applied to both input channels. Figure 3 As shown in the ef section, the output pulse frequency is highest (>99%) when ΔPhase is close to 0° (i.e., the time difference is close to 0); when ΔPhase increases to 90°, the output is suppressed. This indicates that the detector is highly sensitive to the time coincidence of the input signal.

[0043] Example 4: Bionic Auditory Localization System

[0044] like Figure 4 As shown, refer to Figure 4 Part b describes the construction of a biomimetic auditory localization system containing five overlapping detectors. Two microphones (spaced d = 20 cm) simulate binaural hearing, converting sound signals into electrical signals. The delay lines of the five detectors are pre-calibrated to compensate for different ITDs (e.g., ITD1 = 500 μs, ITD3 = 100 μs, ITD4 = -294 μs).

[0045] When the sound source comes from different directions (corresponding to different ITDs), only the detector whose ITD is precisely compensated by a specific delay line will be activated. Figure 4 (Part c). Experiments have shown that the system can operate stably in noisy environments and achieve accurate sound source localization.

[0046] The main advantages of this invention are: 1. Modularity and scalability: The single neuron unit has a simple structure and can be flexibly connected or combined into complex networks as basic modules, making it easy to integrate on a large scale. 2. High temporal accuracy: The time resolution of pulse propagation and coincidence detection reaches the microsecond level, meeting the high requirements of biological nervous systems for temporal information processing. 3. High biosimilarity: It successfully simulates key neurobiological characteristics such as pulse firing, refractory period, unidirectional propagation, and coincidence detection. 4. Strong applicability: The constructed biomimetic auditory system verifies its practical value and robustness in real-time signal processing (such as sound source localization). 5. Hardware-friendly: Based on readily available circuit components, it can be further integrated with micro / nano-scale memristors, laying the foundation for the realization of high-performance neuromorphic chips.

[0047] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The above descriptions are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A biomimetic auditory localization system based on memristor circuits and constructed from bridging artificial neuron units, characterized in that, include: A pair of sound signal acquisition modules are used to simulate sound signals from the left and right ears and convert them into electrical signals of a pulse sequence. The pulse sequence is received and propagated by a pulse propagation network, which introduces a specific delay to compensate for a specific binaural time difference. The pulse propagation network is composed of cascaded bridging artificial neuron units for attenuation-free unidirectional propagation of the pulse signal, facilitating the simulation of signal conduction on biological axons. The input of the coincidence detector is connected to the output of the corresponding pulse propagation network. When the two delayed pulse signals coincide in time, the output unit of the corresponding coincidence detector generates an output pulse to achieve sound source localization. The pulse propagation network is composed of at least two bridging artificial neuron units connected in series; wherein, the output terminal of the second bridging capacitor of the preceding bridging artificial neuron unit is connected to the input terminal of the following bridging artificial neuron unit, so that the pulse signal propagates unidirectionally along the series direction of the at least two bridging artificial neuron units. The overlap detector includes one output unit and two input units, with the output terminals of the input units bridged to the input terminals of the output units. The input terminals of the overlap detector's input units are bridged to the output terminals of the pulse propagation network to receive delayed multiple signals. When the multiple signals overlap in time, the output channel of the overlap detector generates an output pulse. The bridging artificial neuron unit includes: ion channels simulating biological neurons via two branches, each branch including a memristor simulation circuit, a capacitor, and a load resistor; a bridging capacitor connecting the two branches; and the bridging artificial neuron unit being connected to an external circuit via another bridging capacitor.

2. The bionic auditory positioning system according to claim 1, characterized in that, The first memristor analog circuit M1 with threshold switching characteristics has a first capacitor Cm1 connected in parallel across its two ends and a first load resistor Rm1 connected in series, forming the first branch of the simulated sodium ion channel. The second memristor analog circuit M2, which has threshold switching characteristics, has a second capacitor Cm2 connected in parallel across its two ends and a second load resistor Rm2 connected in series, forming the second branch of the simulated potassium ion channel. The first bridging capacitor Cb1 is connected between the first branch and the second branch; The second bridging capacitor Cb2 is connected between the first branch and / or the second branch and the external circuit.

3. The bionic auditory positioning system according to claim 1, characterized in that, The structure types of the pulse propagation network include chain structure, tree branch structure and two-dimensional grid structure.

4. The bionic auditory positioning system according to claim 1, characterized in that, Also includes: The weights and time integration windows of each input signal are adjusted by adjusting the load resistor Rm1 parameter in the output unit and / or the second bridge capacitor Cb2 parameter in the input unit.

Images