A smoke machine infrared signal anti-interference method

Abstract

The application discloses a kind of smoke machine infrared signal anti-interference methods, comprising the following steps: S100, signal pre-processing and spatial feature extraction: frequency band screening is carried out to infrared receiving signal by dynamic band-pass filter circuit, to carry out out-of-band noise suppression, while the spatial difference characteristics of signal are obtained using double-channel differential sampling circuit;S200, feature enhancement blind source separation: the mixed signal after pre-processing is decomposed based on blind source separation algorithm, in the decomposition process, embedding joint feature constraint of time domain, frequency domain and space domain, eliminate interference and extract multi-dimensional feature parameters;S300, interference level decision and adaptive response: the feature parameters are classified by lightweight deep learning model, identify gesture signal and interference signal, and execute adaptive strategy according to interference level;The application can dynamically adapt to complex interference environment, solves the problem that fixed frequency band filtering cannot dynamically adapt to interference, improves signal stability, and guarantees the real-time performance of user operation.

Description

Technical Field

[0001] This invention belongs to the field of smart home control technology, specifically relating to a method for preventing interference with infrared signals from a range hood. Background Technology

[0002] In smart home scenarios, infrared sensors are widely used in applications such as gesture control for range hoods and obstacle avoidance for robotic vacuum cleaners due to their low cost and fast response. However, infrared frequency resources are limited, and infrared sensors from different devices may interfere with each other due to overlapping or similar operating frequencies, leading to false triggering or malfunction of the devices.

[0003] Taking range hoods and robot vacuums as examples: The active infrared gesture sensing module of a range hood emits infrared light through a transmitter. When a hand is waved at close range (5-12cm), the directional reflected light triggers a signal from the receiver. Meanwhile, the infrared obstacle avoidance sensor of a robot vacuum detects obstacles using diffuse reflection, scattering its emitted infrared light in all directions. When both operate simultaneously, two types of interference problems may occur: First, the infrared signal emitted by the robot vacuum may be misinterpreted as a gesture by the range hood, leading to accidental activation or function switching; second, the infrared signal from the range hood may be misjudged as an obstacle by the robot vacuum, causing path planning confusion or abnormal shutdown.

[0004] Existing infrared signal anti-interference technologies mostly rely on fixed-band filtering or simple signal threshold judgment. However, fixed-band filtering cannot dynamically adapt to complex interference environments, and fixed-frequency infrared emission is prone to continuous interference with other devices. Furthermore, relying solely on software algorithms for filtering can increase signal processing latency (e.g., IIR / FIR filters require point-by-point calculations), impacting user experience. With the increasing number of smart home devices, the lack of unified standards for infrared sensor frequency band planning across different brands and models exacerbates interference problems, urgently requiring a solution. Summary of the Invention

[0005] This invention provides a method for preventing interference with infrared signals in a smoke machine, aiming to improve the existing infrared signal interference prevention technologies, which cannot dynamically adapt to complex interference environments and easily lead to increased signal processing delays.

[0006] This invention is implemented as follows. A method for preventing interference with infrared signals from a range hood includes the following steps: S100, Signal Preprocessing and Spatial Feature Extraction: The infrared received signal is filtered by a dynamic bandpass filter circuit to suppress out-of-band noise, while the spatial difference features of the signal are obtained by a dual-channel differential sampling circuit. S200, Feature-enhanced blind source separation: Based on the blind source separation algorithm, the preprocessed mixed signal is decomposed. During the decomposition process, joint feature constraints of time domain, frequency domain and spatial domain are embedded to remove interference and extract multi-dimensional feature parameters. S300, Interference Level Decision and Adaptive Response: A lightweight deep learning model is used to classify feature parameters, identify gesture signals and interference signals, and execute the following strategies based on the interference level: If there is no interference, the range hood will operate in normal mode; For minor interference, use software filtering algorithms to suppress noise; In the event of severe interference, a hardware-level frequency transition will be triggered to a preset backup frequency. The dual-channel differential sampling circuit is equipped with two infrared receiver tubes with a distance of 5-15cm between them. The dual-channel differential sampling circuit adopts the ADC dual-channel synchronous sampling mode and obtains the spatial difference characteristics by calculating the cross-correlation coefficient of the two channel signals. In step S300, if the feature matching degree is ≥90% and the subband energy ratio of 38kHz±2kHz is ≥60%, it is determined to be no interference; if the feature matching degree is <70% or the subband energy ratio is <40%, it is determined to be severe interference, and others are mild interference; wherein, the feature matching degree is the Softmax probability value output by the lightweight deep learning model, and the subband energy ratio is based on the calculation result of STM32 hardware FFT; when it is determined to be severe interference, the hardware-level frequency jumps to the preset backup frequency and broadcasts the current working frequency through BLE to negotiate the frequency band with other smart home devices with infrared function. When the negotiation is abnormal, the abnormal recovery mechanism of hardware watchdog (WWDG) and software heartbeat packet double verification is triggered; in the abnormal state, the current working frequency and interference level are automatically saved to the Flash memory, and the parameter recovery is completed within 100ms after reset.

[0007] Furthermore, the dynamic bandpass filter circuit has a numerically controlled LC filter and a switched capacitor filter. The dynamic bandpass filter circuit achieves dynamic adjustment of the center frequency from 36 to 40 kHz through the numerically controlled LC filter and achieves precise control of the bandwidth of ±1 kHz through the switched capacitor filter.

[0008] Furthermore, in step S200, the blind source separation algorithm includes a preprocessing process, which includes signal centering and whitening. The whitening process is used to remove the correlation between signal dimensions and is accelerated based on the ARMCMSIS-DSP library. The iterative calculation of the blind source separation adopts fixed-point Q15 format operation and the number of iterations is limited to ≤10 times, so that the signal processing delay is <5ms.

[0009] Furthermore, the mixed-signal model is represented as X = AS + N, where: X is the observation signal matrix (m×n), which consists of infrared signals output from a dual-channel differential sampling circuit; A is an unknown mixing matrix (m×k), representing the signal mixing method; S is an unknown source signal matrix (k×n), which contains at least gesture signal and interference signal components; N is a noise matrix that satisfies the following conditions: the number of sensors m ≥ the number of source signals k, and only one Gaussian source signal is allowed to exist.

[0010] Furthermore, the joint feature parameters in the time domain, frequency domain, and spatial domain include: Time domain: signal zero-crossing rate, pulse duty cycle; Frequency domain: 38kHz±2kHz subband energy extracted via STM32 hardware FFT; Spatial domain: Cross-correlation coefficient of dual-channel signals.

[0011] Furthermore, in step S200, interference is eliminated through the following mechanism: S210. Calculate the time-frequency-space characteristic parameters of each decomposed component and evaluate the matching degree with the preset gesture signal feature template. S220. If the characteristic parameters of a certain component deviate from the template threshold range, it is determined to be interference and removed. S230. Reconstruct the filtered components into a pure gesture signal and extract its multidimensional feature parameters.

[0012] Furthermore, the lightweight deep learning model is a 1D-CNN+BiLSTM architecture, with inputs including a 100ms time-domain waveform and 13-dimensional MFCC features. The model is quantized to 8-bit integers using TensorFlowLite, with a Flash memory footprint of <50KB, and inference is accelerated using the AI ​​coprocessor of ESP32-S3.

[0013] The infrared signal anti-interference method for range hoods provided by this invention can dynamically adapt to complex interference environments. It avoids interference signals through dynamic bandpass filtering and hardware-level frequency switching, solving the problem that fixed-band filtering cannot dynamically adapt to interference and improving signal stability. Simultaneously, it employs multi-dimensional feature fusion and a lightweight model to accurately identify gesture signals and interference signals while avoiding the latency problem of traditional software filtering, ensuring real-time operation for users. Other advantages of this invention are described in the following description. Attached Figure Description

[0014] Figure 1 This is a flowchart of the infrared signal anti-interference method for a range hood provided in this embodiment of the invention; Figure 2 This is a flowchart of the interference removal mechanism in step S200. Detailed Implementation

[0015] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0016] The following description, in conjunction with the accompanying drawings and specific embodiments, provides further details: The main hardware components of the infrared anti-interference system for the range hood include: The dual-channel differential sampling circuit has two infrared receiver tubes with a distance of 10cm between them. Dynamic bandpass filter circuit: integrates a numerically controlled LC filter (model LTC1562) and a switched capacitor filter (model MAX7401) to achieve dynamic filtering from 36 to 40 kHz; Main control chip: ESP32-S3-WROOM-1 module, with built-in 240MHz main frequency processor and AI coprocessor, supporting TensorFlowLite model inference; Storage unit: Configured with 4MB Flash memory for storing model parameters and system logs; BLE communication module: Integrates the built-in BLE5.0 function of ESP32-S3, which can realize frequency band negotiation between devices.

[0017] As shown in Figure 1, a method for preventing interference with infrared signals from a range hood includes the following steps: S100. Signal Preprocessing and Spatial Feature Extraction: A dynamic bandpass filter circuit is used to filter the frequency band of the received infrared signal for out-of-band noise suppression. The dynamic bandpass filter circuit includes a numerically controlled LC filter and a switched-capacitor filter. The numerically controlled LC filter achieves dynamic adjustment of the center frequency from 36-40kHz, while the switched-capacitor filter achieves precise bandwidth control of ±1kHz. Simultaneously, a dual-channel differential sampling circuit is used with a dual-channel synchronous sampling mode of an ADC to simultaneously acquire two signals at a sampling rate of 1MHz. These signals are then converted to digital signals by the ADC, and the cross-correlation coefficients of the two channels are calculated to obtain spatial difference features.

[0018] S200, Feature-enhanced blind source separation: The preprocessed mixed signal enters the blind source separation stage. Based on the blind source separation algorithm, the preprocessed mixed signal is decomposed. During the decomposition process, joint feature constraints of the time domain, frequency domain and spatial domain are embedded to remove interference and extract multi-dimensional feature parameters.

[0019] The blind source separation algorithm includes a preprocessing step, which involves signal centering and whitening. First, signal centering is performed to achieve a mean of 0. Then, whitening is used to remove correlations between signal dimensions. This process is accelerated using the ARMCMSIS-DSP library. The iterative computation for blind source separation uses a fixed-point Q15 format, with the number of iterations limited to ≤10 to ensure a signal processing latency of <5ms. The mixed signal model is represented as X=AS+N, where: X is the observed signal matrix (m×n), composed of infrared signals output from a dual-channel differential sampling circuit; A is the unknown mixing matrix (m×k), representing the signal mixing method; S is the unknown source signal matrix (k×n), containing at least gesture signals and interference signal components; and N is the noise matrix, satisfying that the number of sensors m ≥ the number of source signals k, and only one Gaussian source signal is allowed.

[0020] The joint characteristic parameters of the time domain, frequency domain, and spatial domain include: Time domain: signal zero-crossing rate, pulse duty cycle. The zero-crossing rate of a normal gesture signal is about 50-80 times / ms, and the typical value of the pulse duty cycle is 30%-70%.

[0021] Frequency domain: 38kHz±2kHz subband energy extracted by STM32 hardware FFT; the energy proportion of normal gesture signals in this frequency band is usually >60%.

[0022] Spatial domain: Calculate the cross-correlation coefficient of the two-channel signals. The correlation coefficient of the gesture signal is usually >0.7.

[0023] like Figure 2 As shown, in step S200, interference is eliminated through the following mechanism: S210. Calculate the time-frequency-space characteristic parameters of each decomposed component and evaluate the matching degree with the preset gesture signal feature template.

[0024] After decomposing the signal into multiple signal components using the blind source separation algorithm, time-domain, frequency-domain, and spatial-domain feature parameters are extracted for each component.

[0025] In the time domain, the zero-crossing rate of the signal components is calculated, i.e., the number of times the signal amplitude crosses the zero axis per unit time, which is achieved by statistically analyzing the positive and negative changes in amplitude between adjacent sampling points. Simultaneously, the pulse duty cycle is calculated, i.e., the proportion of time the signal is in a high-level state within a complete cycle, which is determined by identifying the start and end times of the signal pulse.

[0026] In the frequency domain, the STM32 hardware FFT function is used to perform a Fast Fourier Transform on the signal components to obtain their spectral distribution. The focus is on extracting the energy of the 38kHz±2kHz sub-band, using the sum of squared spectral amplitudes within this band as a measure of the sub-band energy.

[0027] In the spatial domain, based on a dual-channel differential sampling circuit, the cross-correlation coefficients of the two channel signal components are calculated. Using a discrete signal cross-correlation coefficient calculation method, the two-channel signals are first mean-removed, then the covariance and standard deviation are calculated to obtain the cross-correlation coefficients, thus reflecting the spatial consistency of the signals.

[0028] The preset gesture signal feature template is obtained through statistical analysis of the feature parameters of a large number of normal gesture signal samples. Each feature parameter has a corresponding standard range, such as a zero-crossing rate standard range of 50-80 times / second, a pulse duty cycle standard range of 30%-70%, a 38kHz±2kHz sub-band energy standard range of more than 60% of the total energy, and a dual-channel cross-correlation coefficient standard range of more than 0.7. The extracted feature parameters of each decomposed component are compared with the preset template, and the deviation between the feature parameters and the template is calculated using Euclidean distance or cosine similarity to evaluate the matching degree.

[0029] S220. If the characteristic parameters of a certain component deviate from the template threshold range, it is determined to be interference and is removed.

[0030] S230. Reconstruct the filtered components into a clean gesture signal and extract its multidimensional feature parameters. Reconstruct the retained signal components after filtering using the reverse process of the blind source separation algorithm to recover the clean gesture signal. Extract time-domain, frequency-domain, and spatial-domain multidimensional feature parameters from the reconstructed clean gesture signal again. These parameters will serve as input data for subsequent lightweight deep learning models, used for further interference level decisions and adaptive responses. When extracting multidimensional feature parameters, the same calculation methods and standards as in step S210 are used to ensure the consistency and accuracy of the feature parameters. S300, Interference Level Decision and Adaptive Response: A lightweight deep learning model is used to classify feature parameters and identify gesture signals and interference signals. The lightweight deep learning model is a 1D-CNN+BiLSTM architecture. The input includes a 100ms time-domain waveform and 13-dimensional MFCC features. The model is quantized to 8-bit integers using TensorFlowLite, with a Flash footprint of <50KB, and inference is accelerated using the AI ​​coprocessor of the ESP32-S3.

[0031] If the feature matching degree is ≥90% and the sub-band energy ratio of 38kHz±2kHz is ≥60%, it is considered interference-free, and the appliance operates in normal mode. If the feature matching degree is <70% or the sub-band energy ratio is <40%, it is considered severe interference, triggering a hardware-level frequency jump to a preset backup frequency. Other conditions are considered mild interference, and software filtering algorithms are used to suppress noise. The feature matching degree is the Softmax probability value output by the lightweight deep learning model, and the sub-band energy ratio is based on the calculation result of the STM32 hardware FFT. For example, if the model output matching degree is 92% and the sub-band energy ratio is 65%, it is considered interference-free, and the appliance responds normally to gesture commands. Conversely, if the matching degree is 75% and the sub-band energy ratio is 55%, it is considered mild interference, and the system uses a Kalman filter algorithm to suppress noise. When severe interference is detected, the hardware-level frequency switches to a preset backup frequency while simultaneously broadcasting the current operating frequency via BLE to negotiate frequency bands with other smart home devices that have infrared functionality. If the negotiation fails, such as failing to receive a response three times consecutively, a dual-verification mechanism using a hardware watchdog timer (WWDG) and software heartbeat is triggered to ensure normal program operation. In abnormal situations, environmental parameters such as the current operating frequency and interference level are automatically saved to Flash memory. After a reset, this data can be read to quickly restore the system to its pre-abnormal state within 100ms, completing parameter recovery.

[0032] S400 and the range hood generate the following output commands based on the interference level: Mild Interference Handling: The carrier frequency is fine-tuned by ±200Hz. By adjusting the carrier frequency, the center frequency of interfering signals can be avoided, achieving frequency domain isolation. Simultaneously, a slow-flashing yellow LED is controlled to visually inform the user of the presence of mild interference and that the system has automatically adjusted its frequency. The flashing frequency is set to 1 time per second, ensuring uninterrupted user operation while continuously alerting the user to environmental anomalies.

[0033] Severe Interference Handling: The infrared receiver is disabled for 500ms to cut off interference input and reset the signal link. A strong vibration alert is triggered, for example, by generating a strong vibration at a frequency of 2Hz and an amplitude of 0.5mm through the internal vibration motor of the range hood (such as an eccentric wheel motor), lasting for 2-3 seconds. Compared to simple light alerts, vibration feedback is more likely to attract user attention, especially in noisy kitchen environments. Furthermore, cloud alarms are reported for remote monitoring and facilitate backend analysis by maintenance personnel. If the severe interference reporting rate for a certain model of range hood exceeds a threshold in a specific area (e.g., 10% of users / week), the system automatically triggers an OTA upgrade, pushing optimized feature templates or frequency switching strategies.

[0034] In the four main steps of the anti-interference method for infrared signals of smoke hoods provided in this application, step S100 initially filters the frequency band and obtains spatial features through hardware means; step S200 performs deep separation and feature extraction at the algorithm level based on the results of step S100; step S300 makes intelligent decisions based on the features extracted in step S200 to achieve graded response; and step S400 is further execution and feedback based on the interference level decision results of step S300. Through hardware actions, user interaction and system linkage, the abstract level judgment is transformed into specific anti-interference execution instructions, forming a complete "detection-analysis-decision-execution" closed loop.

[0035] In summary, the anti-interference method for infrared signals of a smoke machine provided in this application can dynamically adapt to complex interference environments. By using dynamic bandpass filtering and hardware-level frequency switching to avoid interference signals, it solves the problem that fixed-band filtering cannot dynamically adapt to interference, thus improving signal stability. Simultaneously, by employing multi-dimensional feature fusion and a lightweight model, it accurately identifies gesture signals and interference signals while avoiding the latency issues of traditional software filtering, ensuring the real-time performance of user operations.

[0036] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for preventing interference with infrared signals from a range hood, characterized in that, Includes the following steps: S100, Signal Preprocessing and Spatial Feature Extraction: The infrared received signal is filtered by a dynamic bandpass filter circuit to suppress out-of-band noise, while the spatial difference features of the signal are obtained by a dual-channel differential sampling circuit. S200, Feature-enhanced blind source separation: Based on the blind source separation algorithm, the preprocessed mixed signal is decomposed. During the decomposition process, joint feature constraints of time domain, frequency domain and spatial domain are embedded to remove interference and extract multi-dimensional feature parameters. S300, Interference Level Decision and Adaptive Response: A lightweight deep learning model is used to classify feature parameters, identify gesture signals and interference signals, and execute the following strategies based on the interference level: If there is no interference, the range hood will operate in normal mode; For minor interference, use software filtering algorithms to suppress noise; In the event of severe interference, a hardware-level frequency transition will be triggered to a preset backup frequency. The dual-channel differential sampling circuit is equipped with two infrared receiver tubes with a distance of 5-15cm between them. The dual-channel differential sampling circuit adopts the ADC dual-channel synchronous sampling mode and obtains the spatial difference characteristics by calculating the cross-correlation coefficient of the two channel signals. In step S300, if the feature matching degree is ≥90% and the subband energy ratio of 38kHz±2kHz is ≥60%, it is determined to be no interference; if the feature matching degree is <70% or the subband energy ratio is <40%, it is determined to be severe interference, and others are mild interference; wherein, the feature matching degree is the Softmax probability value output by the lightweight deep learning model, and the subband energy ratio is based on the calculation result of STM32 hardware FFT; when it is determined to be severe interference, the hardware-level frequency jumps to the preset backup frequency and broadcasts the current working frequency through BLE to negotiate the frequency band with other smart home devices with infrared function. When the negotiation is abnormal, the abnormal recovery mechanism of hardware watchdog (WWDG) and software heartbeat packet double verification is triggered; in the abnormal state, the current working frequency and interference level are automatically saved to the Flash memory, and the parameter recovery is completed within 100ms after reset.

2. The method for preventing interference with infrared signals of a range hood according to claim 1, characterized in that, The dynamic bandpass filter circuit has a numerically controlled LC filter and a switched capacitor filter. The dynamic bandpass filter circuit achieves dynamic adjustment of the center frequency from 36 to 40 kHz through the numerically controlled LC filter and achieves precise control of the bandwidth of ±1 kHz through the switched capacitor filter.

3. The method for preventing interference with infrared signals from a smoke machine according to claim 1, characterized in that, In step S200, the blind source separation algorithm includes a preprocessing process, which includes signal centering and whitening. The whitening process is used to remove the correlation between signal dimensions and is accelerated based on the ARMCMSIS-DSP library. The iterative calculation of the blind source separation adopts fixed-point Q15 format operation and the number of iterations is limited to ≤10 times, so that the signal processing delay is <5ms.

4. The method for preventing interference with infrared signals of a smoke machine according to claim 3, characterized in that, The mixed-signal model is represented as X = AS + N, where: X is the observation signal matrix (m×n), which consists of infrared signals output from a dual-channel differential sampling circuit; A is an unknown mixing matrix (m×k), representing the signal mixing method; S is an unknown source signal matrix (k×n), which contains at least gesture signal and interference signal components; N is a noise matrix that satisfies the following conditions: the number of sensors m ≥ the number of source signals k, and only one Gaussian source signal is allowed to exist.

5. The method for preventing interference with infrared signals of a range hood according to claim 4, characterized in that, The joint feature parameters of the time domain, frequency domain, and spatial domain include: Time domain: signal zero-crossing rate, pulse duty cycle; Frequency domain: 38kHz±2kHz subband energy extracted via STM32 hardware FFT; Spatial domain: Cross-correlation coefficient of dual-channel signals.

6. The method for preventing interference with infrared signals of a range hood according to claim 5, characterized in that, In step S200, interference is eliminated through the following mechanism: S210. Calculate the time-frequency-space characteristic parameters of each decomposed component and evaluate the matching degree with the preset gesture signal feature template. S220. If the characteristic parameters of a certain component deviate from the template threshold range, it is determined to be interference and removed. S230. Reconstruct the filtered components into a pure gesture signal and extract its multidimensional feature parameters.

7. The method for preventing interference with infrared signals of a range hood according to claim 6, characterized in that, The lightweight deep learning model is a 1D-CNN+BiLSTM architecture. The input includes a 100ms time-domain waveform and 13-dimensional MFCC features. The model is quantized to 8-bit integers using TensorFlowLite, with a Flash memory footprint of <50KB. The AI ​​coprocessor of ESP32-S3 is used to accelerate inference.

Images