Reservoir calculation method and system for analog signal processing
By converting the analog signal into optical pulse signals and inputting them to the photomemristor array, and using its high-dimensional mapping characteristics for processing, the problems of high calculation cost and poor real-time performance of traditional analog signal processing methods are solved, and efficient and accurate Analog signal processing.
Patent Information
- Application Number
- CN202410974680.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-01
AI Technical Summary
Traditional analog signal processing methods are difficult to directly apply to analog timing signals, and traditional reserve pool computing is implemented based on software, with high calculation costs and difficult to directly apply in hardware. How to achieve continuous input and efficient processing of analog signals is a challenge.
By collecting analog signals, pre-processing and digitizing, converting them into optical pulse signals, inputting them to photomemristor arrays, and signal processing is performed using their high-dimensional mapping characteristics, breaking the linear limitations of traditional reserve pool calculations and reducing calculation costs.
It realizes efficient and accurate processing of analog signals, improves real-time and efficiency, reduces calculation costs, and significantly simplifies the signal processing process.
Smart Images

Figure CN118885752A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of analog signal processing, and in particular to the technical field of a reserve pool calculation method and system for analog signal processing. Background Art
[0002] In the field of analog signal processing, traditional methods face many challenges, especially in processing time series signals such as speech signals and electromyographic signals. First, these signals are highly continuous and complex, making it difficult to directly apply traditional digital signal processing methods to analog signal processing, resulting in unsatisfactory processing results. Secondly, although recursive neural networks (RNNs) have significant advantages in processing time series signals, their training process is complex and requires a lot of computing resources and time, which limits their widespread use in practical applications.
[0003] In order to solve the above problems, researchers began to explore methods based on reservoir computing. Reservoir computing uses a randomly generated and fixed "reservoir" to process information, and only needs to train the output weights of the network, which greatly simplifies the training process. However, traditional reservoir computing is mainly based on software implementation, with high computational costs and difficulty in direct application to hardware, which limits its performance in terms of real-time and efficiency.
[0004] At the same time, dynamic memristors, as an emerging non-volatile memory, have high-dimensional mapping capabilities and can exhibit plasticity similar to biological synapses. Their physical properties directly simulate the dynamic behavior of neural networks in hardware; specifically, photoelectric memristors can control their resistance states through voltage and light, providing new possibilities for reservoir calculations of analog signals. However, there are still some difficulties when directly applying photoelectric memristors to analog timing signal processing. First, analog signals are continuous, while the input of photoelectric memristors is usually a discrete pulse signal. How to achieve continuous input of analog signals is a problem that needs to be solved. Secondly, the amplitude range of analog signals is large. How to map them into the response range of photoelectric memristors to ensure the accuracy and effectiveness of signal processing is also a challenge. Summary of the invention
[0005] The purpose of the present invention is to overcome the above-mentioned defects in the prior art and to provide a reserve pool calculation method and system for analog signal processing.
[0006] To achieve the above object, the technical solution of the present invention is as follows:
[0007] The present invention provides a reserve pool calculation method for analog signal processing, comprising:
[0008] S1, collect the analog signal to be processed;
[0009] S2, preprocessing the analog signal to obtain the preprocessed analog signal, and making the amplitude of the preprocessed analog signal within the range of the relaxation time of the photoelectric memristor array;
[0010] S3, digitizing the processed analog signal to obtain a pulse signal;
[0011] S4, converting the pulse signal using an optical modulator or a laser to obtain an optical pulse signal;
[0012] S5, inputting the optical pulse signal into the photoelectric memristor array, the photoelectric memristor array realizes high-dimensional mapping of the signal, obtains a mapping result, and outputs the mapping result.
[0013] Optionally, the analog signal is a speech signal or an electromyographic signal.
[0014] Optionally, the specific steps of preprocessing the analog signal in S2 are as follows:
[0015] S21: performing signal screening on the analog signal to obtain the analog signal after signal screening;
[0016] S22: Determine the relaxation time range of the photoelectric memristor array;
[0017] S23: Performing data conversion on the analog signal after signal screening, and converting it into the relaxation time range of the photoelectric memristor array.
[0018] Optionally, the data conversion in S23 includes linear transformation or normalization.
[0019] Optionally, the signal screening of S21 includes spectrum shaping or baseline drift removal.
[0020] Optionally, the steps of the digital processing of S3 are as follows:
[0021] S31, using a sampler to sample the processed analog signal, and converting the continuous analog signal into a plurality of discrete pulse width values;
[0022] S32. Use an optical modulator or a laser to convert all pulse width values into optical pulse signals.
[0023] Optionally, the photoelectric memristor array is composed of a plurality of photoelectric memristor units, and the photoelectric memristor units simulate the dynamic behavior of neurons in a neural network by changing their resistance states.
[0024] Optionally, the photoelectric memristor unit includes a plurality of photosensitive elements.
[0025] A system for any one of the above-mentioned reserve pool calculation methods for analog signal processing, wherein the system comprises:
[0026] The photoelectric memristor array;
[0027] A signal acquisition module, used for acquiring analog signals;
[0028] A signal processing module, the signal processing module is electrically connected to the signal acquisition module and is used to pre-process and digitize the analog signal;
[0029] A signal mapping module, the signal mapping module is electrically connected to the signal processing module, and is used to output the pulse signal as an optical pulse signal, and to give the photoelectric memristor array an optical pulse stimulus;
[0030] An output module is electrically connected to the photoelectric memristor array and is used to receive the signal after high-dimensional mapping of the photoelectric memristor array and perform subsequent output.
[0031] A system for a reserve pool calculation method for analog signal processing, further comprising: a feedback module, one end of which is electrically connected to the signal processing module, and the other end of which is electrically connected to the output module; the feedback module can adjust the parameters of the signal processing module according to the output result of the output module to optimize the signal processing effect.
[0032] It can be seen from the above technical solution that the present invention simulates the reserve pool calculation by the nonlinear mapping characteristics of the photoelectric memristor array, thereby realizing high-dimensional mapping of analog signals, breaking the original limitation that the reserve pool calculation can only be used to process digital signals, reducing the multiple calculation steps that may be required in traditional linear digital systems, simplifying the signal processing process, and thus significantly reducing the computing cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 A flow chart of a method for calculating a reserve pool for analog signal processing according to the present invention;
[0034] Figure 2 A schematic diagram of a system for a reserve pool calculation method for analog signal processing according to the present invention;
[0035] Figure 3 A schematic diagram of signal conversion of a reservoir calculation method for analog signal processing according to the present invention. DETAILED DESCRIPTION
[0036] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.
[0037] For questions such as how to achieve continuous input of analog signals to photoelectric memristors, and how to map analog signals to the response range of photoelectric memristors when the amplitude range is large to ensure the accuracy and effectiveness of signal processing, see Figure 1 and Figure 3 The present invention provides a reserve pool calculation method for analog signal processing, comprising:
[0038] S1, collect the analog signal to be processed;
[0039] S2, preprocessing the analog signal to obtain the processed analog signal, and making the amplitude of the preprocessed analog signal within the range of the relaxation time of the photoelectric memristor array
[0040] S3, digitizing the processed analog signal to obtain a pulse signal;
[0041] S4, converting the pulse signal using an optical modulator or a laser to obtain an optical pulse signal;
[0042] S5, inputting the optical pulse signal into the photoelectric memristor array, the photoelectric memristor array realizes high-dimensional mapping of the signal, and obtains a mapping result; and outputting the mapping result.
[0043] The present invention can optimize the signal processing process by collecting the analog signal to be collected and processing the analog signal to within the range of relaxation time, so that the photoelectric memristor array can respond to the input signal more effectively and improve the accuracy and efficiency of signal processing; the reason is that if the amplitude of the analog signal exceeds the processable range of the photoelectric memristor array, the photoelectric memristor array may be saturated or overloaded. Saturation will cause the photoelectric memristor array to be unable to accurately record the change of the signal, and overload may damage the photoelectric memristor array. By limiting the signal amplitude to the relaxation time range, this situation can be avoided, the memristor is protected and its long-term stability is ensured. In addition, the present invention can quickly interact with the photoelectric memristor array to achieve rapid signal processing by converting the pulse signal into an optical pulse signal using an optical modulator or a laser, breaking the limitation that the original reserve pool calculation can only be used to process digital signals; in addition, the physical characteristics of the photoelectric memristor array of the present invention can directly simulate the dynamic behavior of the neural network in hardware, that is, by directly realizing the reserve pool calculation by itself, the calculation cost can be significantly reduced.
[0044] In some embodiments, the photoelectric memristor array may be provided with an output layer, and the output layer generally refers to an electrode connected to the photoelectric memristor array, and is used to output the obtained mapping result.
[0045] In some embodiments, the analog signal may be a speech signal or an electromyographic signal; for a speech signal, the information it contains may include speech speed, intonation, volume, and pronunciation. When processing speech signals, these features need to be accurately extracted to achieve speech recognition, speech synthesis, or other related applications. Due to the fast response characteristics of the photoelectric memristor array, this method has significant advantages and application prospects in speech signal processing situations with high real-time requirements. Electromyography (EMG) is an electrical signal generated by muscle contraction, which is usually used to diagnose muscle diseases, evaluate muscle function, or control prostheses. Electromyographic signals are also continuous analog signals, and efficient and accurate signal processing methods are required to extract useful information from them. This analog signal processing method based on the photoelectric memristor array can not only improve the efficiency and accuracy of signal processing, but also provide new ideas and methods for research and application in the fields of speech recognition, electromyographic control, etc.
[0046] In some embodiments, in a reserve pool calculation method for analog signal processing of the present invention, the specific steps of preprocessing the analog signal in S2 are as follows:
[0047] S21: Screening the analog signal to obtain a filtered analog signal; the purpose of this step is to highlight the key features in the analog signal, so as to determine the signal parts that are critical to subsequent processing and analysis. For example, in audio signal processing, a bandpass filter is generally used to screen signals in a specific frequency range. In biomedical signal processing, a high-pass filter is generally used to remove baseline drift in electromyographic signals.
[0048] S22: Determine the relaxation time range of the photoelectric memristor array; In this step, the relaxation time range of the photoelectric memristor array needs to be determined to ensure that the transformation of the analog signal in the subsequent steps can match it.
[0049] S23: converting the analog signal after signal screening into a relaxation time range of the photoelectric memristor array. This step may involve scaling, translation or other forms of transformation of the signal to ensure that the amplitude of the signal matches the response characteristics of the photoelectric memristor array.
[0050] In some embodiments, the signal screening of S21 includes spectrum shaping or baseline drift removal.
[0051] Spectral shaping is a process that generally uses a bandpass filter to allow signals within a certain frequency range to pass through, while blocking signals of other frequencies. This process can selectively extract or emphasize information in a certain frequency band; especially in audio signal processing, bandpass filters can filter signals in a specific frequency range. For example, it is necessary to retain the human voice (usually between a few hundred hertz and a few thousand hertz) and remove noise of other frequencies; therefore, it can be applied to speech signal processing;
[0052] Baseline drift removal is often used to remove low-frequency noise. Since baseline drift is a common noise interference in electromyographic signals, it may be caused by a variety of reasons, such as electrode movement, breathing or other physiological movements. The general method is to use a high-pass filter to allow high-frequency signals to pass through and block low-frequency signals, thereby effectively removing low-frequency noise such as baseline drift.
[0053] In some embodiments, the data conversion of step S23 includes linear transformation or normalization. Linear transformation maps the range of the signal to the relaxation time range of the photoelectric memristor array by adjusting the amplitude and offset of the signal. Normalization is a process of converting a signal to a specific range (such as 0 to 1 or -1 to 1). Normalization can adjust the amplitude of the analog signal to the relaxation time range of the photoelectric memristor array, ensuring that the rate of change and amplitude of the signal match the response characteristics of the photoelectric memristor array. Normalization can eliminate dimensional differences in the signal, allowing signals from different sources or different scales to be compared and combined.
[0054] In some embodiments, the steps of the digital processing of S3 are as follows:
[0055] S31. Use a sampler to sample the processed analog signal and convert the continuous analog signal into multiple discrete pulse width values. In this step, the following method can be used: For example, the sampler can sample the processed analog signal at a certain sampling rate (i.e., the number of samples per second). The sampler measures the amplitude of the analog signal at each sampling point and converts these amplitudes into digital representations. Then, these digital values are used to determine the pulse width of the pulse signal. For example, pulse width modulation (PWM) technology can achieve the above solution. Specifically, by changing the width of the pulse signal (i.e., the pulse duty cycle) to simulate the amplitude change of the original analog signal, the desired analog signal waveform can be simulated, so that the output signal is close to the original analog signal in terms of average effect.
[0056] S32. Use an optical modulator or a laser to convert all pulse width values into optical pulse signals. Specifically, the optical modulator is based on a certain physical effect (such as electro-optical effect, acousto-optical effect, etc.) to reflect the information of the electrical signal by changing a certain parameter of the optical signal (such as intensity, phase, frequency or polarization state). For pulse signals, the optical modulator can adjust the intensity, width or phase of its output optical pulse to match the input pulse signal; the principle of the laser emitting an optical pulse signal is slightly different. Specifically, the laser generates a continuous light beam, and the external modulator connected to the laser controls the output of the laser according to the electrical pulse signal to generate an optical pulse signal corresponding to the pulse signal.
[0057] In some embodiments, the photoelectric memristor array is composed of a plurality of photoelectric memristor units, and the photoelectric memristor unit simulates the dynamic behavior of neurons in a neural network by changing its resistance state. Specifically, the photoelectric memristor unit may have a metal-insulator-metal structure (MIM structure, Metal-Insulator-Metal), in which the insulating layer is a special photoelectric material, and its resistance state may change under the influence of light signals such as light intensity and wavelength. This change in resistance state can be used to simulate the excitation or inhibition state of neurons, thereby realizing the information processing function in the neural network. In the photoelectric memristor array, each memristor unit can be regarded as a "synapse", which connects two neurons (i.e., the input end and the output end). When the light pulse signal (i.e., the input signal) acts on the electrical memristor unit, its resistance state changes, thereby affecting the current or voltage signal at the output end. This change can be regarded as an update of the connection strength (i.e., synaptic weight) between neurons, thereby realizing the learning and memory function of the neural network.
[0058] The high-dimensional mapping of optical pulse signals by photoelectric memristor arrays is achieved through memristor mechanism and nonlinear mapping characteristics. Specifically, when the optical pulse signal is irradiated on the photoelectric memristor, the photons are absorbed and stimulate electron-hole pairs. These carriers move under the action of the electric field, which changes the resistance state of the device, thus enabling the photoelectric memristor to respond to the optical pulse signal. At the same time, photoelectric memristors have nonlinear mapping characteristics, which enable them to simulate complex biological neuron behaviors. The principle of nonlinear mapping is that when an optical pulse signal is input, the resistance state of the photoelectric memristor will change with factors such as the intensity, frequency and duration of the optical pulse, thereby achieving nonlinear mapping of the input optical pulse signal and obtaining a mapping result.
[0059] In some embodiments, the photoelectric memristor unit includes a plurality of photosensitive elements. A photosensitive element is an electronic element that is sensitive to light sources and can realize photoelectric conversion based on the semiconductor photoelectric effect. In the photoelectric memristor unit, the photosensitive element is generally used to detect light signals and convert them into electrical signals, thereby affecting the resistance state of the memristor.
[0060] A system for the above-mentioned reservoir calculation method for analog signal processing, see Figure 2 , the system comprises:
[0061] Photoelectric memristor array; In this system, the photoelectric memristor array is used to receive light pulse stimulation and perform high-dimensional mapping based on the stimulation signal to realize information processing and storage.
[0062] The signal acquisition module is used to collect analog signals; the analog signals may come from various sensors or other signal sources, mainly voice signals and electromyographic signals.
[0063] A signal processing module, the signal processing module is electrically connected to the signal acquisition module; the signal acquisition module transmits the acquired analog signal to the signal processing module; the signal processing module pre-processes and digitizes the analog signal; the signal processing module can adjust the amplitude of the signal to a uniform range, so as to facilitate subsequent mapping to the photoelectric memristor array by the signal mapping module;
[0064] A signal mapping module, the signal mapping module is electrically connected to the signal processing module; the signal processing module transmits the processed pulse signal to the signal mapping module; the signal mapping module is used to output the pulse signal as an optical pulse signal and give the photoelectric memristor array an optical pulse stimulus;
[0065] The output module is electrically connected to the photoelectric memristor array and is used to receive the signal after high-dimensional mapping of the photoelectric memristor array and perform subsequent output.
[0066] A system for the reserve pool calculation method for analog signal processing, in addition to the above modules, can also add a feedback module, one end of the feedback module is electrically connected to the signal processing module, and the other end of the feedback module is electrically connected to the output module; the feedback module can adjust the parameters of the signal processing module according to the output result of the output module to optimize the signal processing effect.
[0067] In summary, the present invention provides a reserve pool calculation method for analog signal processing, which can perform reserve pool calculation based on a photoelectric memristor array; first, the method can screen the collected analog signal, thereby highlighting the key features of the analog signal; then, by determining the relaxation time range of the photoelectric memristor and adjusting the amplitude of the analog signal accordingly, it can ensure that the signal is effectively processed without being saturated or overloaded, which avoids signal distortion and damage to the photoelectric memristor. In addition, by converting the analog signal into an optical pulse signal, the present invention utilizes the memristive characteristics and nonlinear mapping characteristics of the photoelectric memristor to break the limitations of traditional reserve pool calculations and achieve high-speed processing of analog signals. This method not only improves the speed of signal processing, but also reduces the computing cost and improves the overall energy efficiency ratio by directly simulating the dynamic behavior of the neural network at the hardware level.
[0068] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.
Claims
1. A reserve pool calculation method for analog signal processing, characterized in that: include: S1, collect the analog signal to be processed; S2, preprocessing the analog signal to obtain the preprocessed analog signal, and making the amplitude of the preprocessed analog signal within the range of the relaxation time of the photoelectric memristor array; S3, digitizing the processed analog signal to obtain a pulse signal; S4, converting the pulse signal using an optical modulator or a laser to obtain an optical pulse signal; S5, inputting the optical pulse signal into the photoelectric memristor array, the photoelectric memristor array realizes high-dimensional mapping of the signal, obtains a mapping result, and outputs the mapping result.
2. The method for calculating a reserve pool for analog signal processing according to claim 1, characterized in that: The analog signal is a speech signal or an electromyographic signal.
3. The reserve pool calculation method for analog signal processing according to claim 1, characterized in that: The specific steps of preprocessing the analog signal in S2 are as follows: S21: performing signal screening on the analog signal to obtain the analog signal after signal screening; S22: Determine the relaxation time range of the photoelectric memristor array; S23: Performing data conversion on the analog signal after signal screening, and converting it into the relaxation time range of the photoelectric memristor array.
4. The reserve pool calculation method for analog signal processing according to claim 3, characterized in that: The data conversion in S23 includes linear transformation or normalization.
5. The reserve pool calculation method for analog signal processing according to claim 3, characterized in that: The signal screening of S21 includes spectrum shaping or removing baseline drift.
6. The reserve pool calculation method for analog signal processing according to claim 3, characterized in that: The steps of the S3 digital processing are as follows: S31, using a sampler to sample the processed analog signal, and converting the continuous analog signal into a plurality of discrete pulse width values; S32. Use an optical modulator or a laser to convert all pulse width values into optical pulse signals.
7. The reserve pool calculation method for analog signal processing according to claim 1, characterized in that: The photoelectric memristor array is composed of a plurality of photoelectric memristor units, and the photoelectric memristor units simulate the dynamic behavior of neurons in a neural network through changes in their resistance states.
8. The reserve pool calculation method for analog signal processing according to claim 7, characterized in that: The photoelectric memristor unit includes a plurality of photosensitive elements.
9. A system for a reservoir calculation method for analog signal processing according to any one of claims 1 to 8, characterized in that: The system includes: The photoelectric memristor array; A signal acquisition module, used for acquiring analog signals; A signal processing module, the signal processing module is electrically connected to the signal acquisition module and is used to pre-process and digitize the analog signal; A signal mapping module, the signal mapping module is electrically connected to the signal processing module, and is used to output the pulse signal as an optical pulse signal, and to give the photoelectric memristor array an optical pulse stimulus; An output module is electrically connected to the photoelectric memristor array and is used to receive the signal after high-dimensional mapping of the photoelectric memristor array and perform subsequent output.
10. The system according to claim 9, characterized in that The system also includes a feedback module, one end of which is electrically connected to the signal processing module, and the other end of which is electrically connected to the output module; the feedback module can adjust the parameters of the signal processing module according to the output result of the output module to optimize the signal processing effect.
Citation Information
Patent Citations
Reservoir calculation device and data processing method
CN114418078A
Method for realizing calculation of endosensory reserve pool based on photosynaptic device
CN114723000A
Synapse element, reservoir computing device including the sane, and reservoir computing method using the same
JP2023013949A
Cited By
Speech recognition method based on memristor and electronic equipment
CN122337184A
