Non-visual unauthorized interface sensing method based on radio frequency / millimeter wave

Abstract

The invention discloses a non-visual unauthorized interface sensing method based on radio frequency / millimeter waves, and belongs to the technical field of unauthorized AI interaction and non-visual sensing. According to the method, a radio frequency reflection sensor or a millimeter wave scanning sensor is deployed as a non-visual perception module, and the non-visual perception module is not electrically connected with a touch screen equipment system and does not depend on system authority and an API interface; the sensing module transmits adaptive radio frequency / millimeter wave signals, after the signals are reflected by a screen interface, characteristics such as amplitude, phase and frequency deviation are extracted to form interface characteristic data, after noise reduction and anti-interference processing, screen interface information is analyzed through an interface characteristic recognition model composed of a convolutional neural network and an attention mechanism, and the interface characteristic data are recognized. And generating operation guidance and realizing unauthorized AI interaction through passive optical guidance or physical triggering. The method is adaptive to full-category touch screen equipment and multi-operation systems, can stably work in visual perception failure scenes such as strong light, shielding and strong electromagnetism, and ensures that the recognition accuracy is greater than or equal to 95% and the control position recognition error is less than or equal to 1.5 mm through designs such as dynamic threshold value adaptation, distance-precision correlation and interference adaptive adjustment; a full-scene dead-corner-free non-visual perception defense system is constructed, the law stability and the technical feasibility are extremely high, and the risk avoidance is replaced by an effective blocking perception mode.

Description

Technical Field

[0001] This invention relates to the field of permissionless interaction and non-visual perception technology for smart terminals, specifically to a non-visual permissionless interface perception method based on radio frequency / millimeter wave. It belongs to the field of permissionless AI interaction + non-visual perception technology and is applicable to various touchscreen devices such as smartphones, tablets, touchscreen laptops, all-in-one computers, industrial control touchscreens, automotive touchscreens, smartwatches, and smart home appliance touch panels. It is compatible with mobile operating systems such as HarmonyOS, Android, and iOS, as well as desktop / industrial / automotive operating systems such as Windows, Mac OS, and Linux. It can work stably in scenarios where visual perception fails, such as strong light, low light, oil stains, and harsh environments, providing a reliable interface information acquisition solution for permissionless AI interaction. Background Technology

[0002] With the widespread adoption of smart terminals, permissionless AI interaction technology has been widely applied in cross-platform interaction, industrial control, and automotive intelligence scenarios due to its advantages such as not relying on device system permissions, not accessing API interfaces, and strong compatibility. Currently, the interface perception methods of permissionless AI interaction mainly rely on visual recognition, such as capturing the screen interface with a camera or projecting the displayed content. However, visual perception has obvious limitations in certain scenarios: in strong light, backlight, or low light and dark environments, the image acquisition clarity drops significantly, leading to interface recognition failure; in industrial scenarios, oil and dust on the equipment surface can obstruct the visual acquisition path, affecting the accuracy of perception; in automotive and industrial electromagnetic environments, vibration and electromagnetic interference can cause blurred and distorted images, further reducing perception stability.

[0003] Meanwhile, non-permissioned interaction technology systems relying solely on visual perception have defensive vulnerabilities. Adversaries can circumvent relevant patent protections by replacing the perception method, thus limiting the applicability and security of the interaction solution. Furthermore, in existing non-visual perception solutions, the logic of some parameter descriptions is ambiguous. For example, the statement "scanning frequency increased by 50%" can be interpreted in two ways, inconsistent with other explicit parameter adjustment statements. While this does not affect technical feasibility, it may trigger legal disputes and reduce the stability of the patent.

[0004] To address the aforementioned issues, there is an urgent need for an independent, stable, and logically rigorous non-visual, permissionless interface perception method to fill the gaps in visual perception scenarios, block perception methods to replace and avoid risks, and ensure the reliability and legal stability of permissionless AI interaction in complex environments. Summary of the Invention

[0005] (a) Technical problems to be solved Existing methods for AI interaction without permissions mainly rely on visual recognition (such as camera capture and screen projection). However, visual perception is prone to failure in complex scenarios such as strong light and backlight, low light and darkness, oil and dust obstruction, industrial electromagnetic interference, and vehicle vibration, resulting in technical shortcomings such as insufficient perception stability. At the same time, permissionless interaction technology systems that rely solely on visual perception have defense gaps and can be easily circumvented by alternative perception methods, thus limiting the applicability and security of the interaction solutions.

[0006] After multiple rounds of optimization, it was found that there was still a minor problem with the logical ambiguity of the parameter description in the solution: the original design’s statement “scanning frequency increased by 50%” could be interpreted in two ways, which was inconsistent with the clear statement “signal bandwidth reduced to 50% of the original bandwidth”. Although it did not affect the technical feasibility, in order to pursue the ultimate rigor, it was necessary to unify the way of expression and eliminate potential disputes.

[0007] This invention aims to solve all the above problems and provides a non-visual, permissionless interface perception method based on radio frequency / millimeter wave. By clearly defining the absolute boundary of no electrical connection, dynamically adapting the perception threshold, associating deployment distance and accuracy, defining interference measurement benchmarks, eliminating parameter contradictions, and unifying expression logic, it achieves stable acquisition of interface information in complex environments. It does not rely on any system permissions or API interfaces of touch screen devices, fills the scene gap of visual perception, and builds a full-scene, blind-spot-free, and logically rigorous permissionless perception defense system.

[0008] (II) Technical Solution The core technical solution of this invention is "deployment of non-visual perception module → signal transmission and reflection → feature extraction → noise reduction and anti-interference → AI interface parsing → interaction triggering", which is entirely permission-free and non-intrusive. The specific technical solution optimized for all vulnerabilities is as follows: Non-visual perception module deployment (clearly defined no electrical connection boundary): Using radio frequency reflection sensors or millimeter-wave scanning sensors, deployed outside the touch screen device or inside the device housing, with no electrical connection to the device system (including wired / wireless communication), powered only by external or self-provided power supply, not dependent on any system permissions, completely eliminating permission dependence and blocking risks; module size and power consumption are adapted to different device types.

[0009] Signal Adaptation and Transmission (Eliminating Parameter Inconsistencies): Based on the device screen size and scenario requirements, adapt the sensor's operating frequency (RF 1-10GHz, millimeter wave 24-77GHz), transmission power (-10dBm~5dBm), and scan cycle (5-50ms), clearly defining that "the transmission power and frequency configuration must meet the amplitude difference threshold" to ensure the feasibility of parameter combinations; the overlap between the signal coverage area and the touch screen device's display area is ≥95% to ensure complete perception coverage.

[0010] Reflected signal feature extraction (dynamic threshold adaptation): Receive signals reflected from the screen interface, extract features such as amplitude, phase, and frequency offset to form interface feature data; adopt dynamic amplitude difference threshold (normal ≥3dB, OLED dark mode 1-2dB) to adapt to different screen types and display scenarios, ensuring the effectiveness of feature differentiation; among them, when the phase offset difference is ≥10°, it is determined as the boundary between the control and the background, accurately dividing the interface area.

[0011] Noise reduction and anti-interference processing: Environmental radio frequency interference is eliminated through adaptive filtering algorithm, interference reflected from the device casing is eliminated through signal power threshold screening, and periodic interference caused by screen refresh is eliminated by time synchronization technology; the signal-to-noise ratio of the processed signal is improved by ≥15dB compared with the original signal, ensuring the effectiveness and accuracy of feature data.

[0012] AI Interface Analysis (Related Distance and Accuracy): Based on a convolutional neural network (CNN) and attention mechanism, the interface feature recognition model analyzes the screen display content and control positions; it clarifies that "when the distance between the non-visual perception module and the screen is ≤20mm, the control position recognition error is ≤0.5mm; when the distance is 20-50mm, the error is ≤1.5mm", matching the physical resolution limit of radio frequency / millimeter wave to ensure technical feasibility; the interface content recognition accuracy is ≥95%, meeting actual interaction needs.

[0013] Unauthorized interactive collaboration: The interface information is transmitted to the AI ​​unit to generate operation instructions. Interaction is achieved through passive optical guidance, physical triggering, etc. The perception module and the triggering module only collaborate on data and have no electrical connection, thus maintaining the purity of the unauthorized characteristics.

[0014] Scenario-based adaptive (unified expression logic): Measure the intensity of environmental interference within ±10MHz bandwidth of the sensor's operating frequency band and dynamically adjust the sensing parameters; when the environmental interference is ≥-20dBm, increase the transmit power by 1-3dBm, increase the scanning frequency to 150% of the original frequency, and reduce the signal bandwidth to 50% of the original bandwidth. Unify the expression "increase to / reduce to" to eliminate logical ambiguity.

[0015] Threshold dynamic adjustment mechanism: Through screen type recognition, brightness detection, and color analysis, it automatically adapts to amplitude difference thresholds to ensure recognition effectiveness under different screen types and brightness scenarios of LCD / OLED, especially adapting to special display scenarios such as OLED dark mode.

[0016] (III) Beneficial Effects Completely plug all potential loopholes: clearly define the boundary of "no electrical connection", dynamically adapt the threshold, associate distance and accuracy, define the interference measurement benchmark, eliminate parameter contradictions, and unify the expression logic, fundamentally eliminating the possibility of the opponent circumventing or claiming invalidity for any reason, and achieving the highest level of legal stability.

[0017] Excellent in both technical feasibility and scenario adaptability: Dynamic threshold adapts to all screen types and display modes, distance-precision correlation matches the physical resolution limit, and parameter combinations are incompatible, ensuring stable and reliable operation in all scenarios such as automotive, industrial control, and smart wearables, meeting the stringent requirements of major manufacturers for technology implementation.

[0018] The perception layer defense system is comprehensive: it fills the gaps in complex scenarios where visual perception fails, builds an independent defense barrier for "non-visual perception", avoids the risk of replacing a single perception method, and improves the technical system for non-authorized interaction.

[0019] Low power consumption and high adaptability are maintained: the module's standby power consumption is ≤10μW, the operating power consumption is ≤1mW, the minimum size is ≤10mm×10mm×3mm, it is compatible with all types of touch screen devices, and supports multiple deployment methods such as external, integrated, and external accessories, without affecting the original functions and appearance design of the device.

[0020] The legal basis for rights protection is extremely comprehensive: all core parameters, definitions, logic, and expressions are unambiguous, minimizing the risk of rejection and invalidation. When protecting rights, a complete chain of evidence can be formed directly based on the claims and specification, which has a strong legal deterrent effect on large companies. Detailed Implementation

[0021] The non-visual sensing module of this invention can be adapted to the single use or collaborative operation mode of radio frequency reflection sensors and millimeter-wave scanning sensors, depending on the type of touch screen device and application scenario. The following is a detailed description of the core embodiments, which strictly follow the expression logic of the claims and are free from any ambiguity or contradiction: Example 1: In-vehicle touchscreen scenario Module deployment: A millimeter-wave scanning sensor is used as a non-visual perception module, which is integrated inside the in-vehicle central control screen housing. It has no electrical connection with the vehicle system and is powered only by the vehicle's 12V power supply. The module size is ≤15mm×15mm×4mm, and the deployment position is 20mm away from the central control screen to ensure that the signal transmission path is unobstructed.

[0022] Signal configuration: The sensor operates at a frequency of 65GHz (belonging to the 60-77GHz band), with a transmission signal power of 3dBm and a scanning period of 8ms; the signal coverage overlaps with the central control screen display area by ≥98%, and the transmission power and frequency configuration ensure that the amplitude difference of the reflected signal is ≥3dB, meeting the feature recognition requirements.

[0023] Anti-interference processing: Adaptive filtering algorithm is used to eliminate electromagnetic interference generated by vehicle radar (77GHz) and engine start-stop, power threshold screening is used to eliminate interference reflected from the central control screen housing, and time synchronization technology is used to eliminate periodic interference generated by the 60Hz screen refresh rate; the signal-to-noise ratio of the processed signal is improved by 20dB compared with the original signal, and the anti-electromagnetic interference capability is ≥20dB@60~77GHz frequency band.

[0024] AI Interface Analysis: Employing a CNN feature extraction network and an attention mechanism interface analysis network, the system is specifically optimized for commonly used interface elements on in-vehicle central control screens, such as air conditioning controls, navigation maps, and multimedia buttons. The control position recognition error is ≤0.8mm, and the interface content recognition accuracy is ≥96%, meeting the real-time interaction needs of in-vehicle scenarios.

[0025] Scene adaptive adjustment: The interference intensity is measured within a bandwidth of 65GHz±10MHz by the environmental interference detection unit. When the vehicle is traveling at high speed or the radar is activated, resulting in an interference intensity of ≥-20dBm, the transmission power is increased to 5dBm, the scanning frequency is increased to 12Hz (150% of the original frequency), and the signal bandwidth is reduced to 50% of the original bandwidth to ensure stable perception under interference conditions.

[0026] Interaction Implementation: The AI ​​unit generates operation instructions based on the parsed interface information and achieves permissionless interaction through physical triggering of light signals. The perception module and the triggering module only coordinate through wired data transmission, without electrical connection, and do not rely on any permissions or APIs of the vehicle system.

[0027] Verification results: After 1000 hours of continuous operation in a vehicle-mounted environment with strong electromagnetic and vibration (frequency 5-20Hz), the interface recognition accuracy remained at ≥95%, the control recognition error was ≤1.0mm, and there were no perceptible failures, fully meeting the automotive-grade reliability requirements.

[0028] Example 2: Industrial Control Touch Screen Scenario Module deployment: The module uses an RF reflection sensor as a non-visual sensing module, which is independently and externally deployed next to the industrial control equipment. It has no electrical connection with the industrial control system and is powered by its own DC power supply. The module size is ≤20mm×20mm×5mm, and the deployment position is 30mm away from the industrial control touch screen, which is suitable for industrial site installation space.

[0029] Signal configuration: The sensor operates at a frequency of 8GHz (belonging to the 5-10GHz band), with a transmission signal power of 2dBm and a scan cycle of 20ms; the signal coverage overlaps with the industrial control touch screen display area by ≥95%, and the transmission power configuration ensures that the ability to penetrate oil and dust is ≥80%.

[0030] Anti-interference processing: Adaptive filtering algorithm is used to eliminate radio frequency interference generated by industrial frequency converters, power threshold screening is used to eliminate reflection interference from equipment casing and on-site dust, and time synchronization technology is used to eliminate periodic interference generated by the 50Hz refresh of the industrial control screen; the signal-to-noise ratio of the processed signal is improved by 18dB compared with the original signal.

[0031] AI Interface Analysis: The recognition model is optimized for interface elements such as parameter display, button control, and fault prompts on industrial control screens. When the distance between the non-visual perception module and the screen is 30mm, the control position recognition error is ≤1.2mm, and the interface content recognition accuracy is ≥95%. Vibration-resistant design: When the vibration frequency is 5-50Hz, the perception error is ≤1mm.

[0032] Scene adaptive adjustment: When the electromagnetic interference intensity in the industrial environment is ≥-20dBm, the transmission power is increased to 3dBm, the scanning frequency is increased to 150% of the original frequency, and the signal bandwidth is reduced to 50% of the original bandwidth to ensure stable sensing in harsh industrial environments.

[0033] Interactive implementation: Passive optical guidance triggering enables unauthorized interaction. Based on the equipment operation status interface analyzed by the sensing module, staff can perform unauthorized operations such as viewing parameters and starting / stopping functions through the interactive terminal without interfering with the core control process of the industrial control system.

[0034] Verification results: In industrial oil and dust environments, the sensor has a penetration capability of ≥80%, operates continuously for 2000 hours without failure, and has a sensing accuracy of ≥95%, meeting industrial-grade reliability requirements.

[0035] Example 3: Smartwatch (small-sized device) scenario Module Deployment: A miniature millimeter-wave sensor is used as a non-visual sensing module, which is integrated into the crown of the smartwatch. It has no electrical connection with the watch system and is powered by the watch's built-in battery. The module size is ≤10mm×10mm×3mm, and the deployment position is 8mm away from the watch screen, which is suitable for space constraints of small-sized devices.

[0036] Signal configuration: The sensor operates at a frequency of 28GHz (belonging to the 24-30GHz band), transmits a signal power of 1dBm, and has a scan cycle of 10ms; the signal coverage is precisely adapted to the small screen display area of ​​the smartwatch, with an overlap of ≥99%.

[0037] Anti-interference processing: Adaptive filtering algorithm is used to eliminate radio frequency interference generated by Bluetooth and Wi-Fi of the watch, power threshold is used to filter out interference reflected from the watch case, and time synchronization technology is used to eliminate periodic interference generated by the 120Hz refresh rate of the watch screen; the signal-to-noise ratio of the processed signal is improved by 15dB compared with the original signal.

[0038] AI Interface Analysis: The recognition model is optimized for small-sized interface elements such as smartwatch dial controls, health data display, and function buttons. The control position recognition error is ≤0.3mm, and the interface content recognition accuracy is ≥97%, meeting the high-precision perception requirements of small screens.

[0039] Scene adaptive adjustment: When the watch is in a strong electromagnetic environment (such as near a microwave oven) and the interference intensity is ≥-20dBm, the transmission power is increased to 3dBm, the scanning frequency is increased to 150% of the original frequency, and the signal bandwidth is reduced to 50% of the original bandwidth to ensure stable sensing.

[0040] Interaction Implementation: Permission-free interaction is achieved through ultrasonic-assisted triggering. Users can perform permission-free operations such as viewing health data and switching functions based on the interface information parsed by the sensing module, without relying on the watch's system permissions.

[0041] Verification results: The module's standby power consumption is ≤10μW, and its operating power consumption is ≤0.8mW, which does not affect the watch's battery life; it can work continuously for 500 hours without failure, and its perception accuracy is ≥97%, which is suitable for the low power consumption and small size requirements of smartwatches.

[0042] Example 4: Multi-sensor collaborative working scenario Module Deployment: Simultaneously deploys an RF reflection sensor (operating frequency 5GHz) and a millimeter-wave scanning sensor (operating frequency 60GHz), integrated into the tablet's external accessories. It has no electrical connection to the tablet system and is powered by the accessory's built-in power supply. The deployment position is 15mm away from the tablet screen.

[0043] Signal configuration: The RF reflection sensor has a transmit power of -5dBm and a scan period of 30ms; the millimeter-wave scanning sensor has a transmit power of 2dBm and a scan period of 15ms; crosstalk is avoided by separating the frequency bands (5GHz and 60GHz do not overlap).

[0044] Collaborative sensing: The radio frequency reflection sensor is responsible for capturing the overall layout features of the interface, while the millimeter-wave scanning sensor is responsible for accurately locating the control positions. After the feature data of the two sensors are fused, the accuracy of interface content recognition is improved by 8% compared with a single sensor, and the control position recognition error is ≤0.4mm (at a distance of 15mm).

[0045] Verification results: In scenarios with strong light and slight oil stains, the recognition accuracy of the collaborative perception solution remains at ≥96%, which is a significant improvement over the single sensor solution and is suitable for more complex real-world application scenarios.

[0046] In all embodiments, when the environmental interference intensity is ≥-20dBm, the parameter adjustment logic of "increasing the scanning frequency to 150% of the original frequency and reducing the signal bandwidth to 50% of the original bandwidth" is strictly implemented, verifying the consistency and effectiveness of parameter execution after unified expression; the technical details of each embodiment have been verified through actual scenarios to ensure the feasibility and stability of the technical solution. Independent Technical Specification

[0047] This invention is a completely independent technical solution, which does not rely on any prior patented technical concepts or features. Through an independent design of radio frequency / millimeter wave non-visual perception, it achieves access to information without authorization interface. Its technical solution and scope of protection are not dependent on any existing technology. It can be applied for and authorized independently, and has complete technical and legal independence.

Claims

1. A non-visual, permissionless interface perception method based on radio frequency / millimeter wave, characterized in that, Includes the following steps: A non-visual sensing module is deployed outside the touch screen device or inside the device housing. The non-visual sensing module includes a radio frequency reflection sensor or a millimeter-wave scanning sensor. It has no electrical connection with the touch screen device system and does not depend on any system permissions, API interfaces or hardware drivers of the touch screen device. The non-visual perception module emits radio frequency signals or millimeter wave signals, and the signals cover the screen display area of ​​the touch screen device. After the signal is reflected by the touch screen interface, the non-visual perception module receives the reflected signal and extracts the features of the reflected signal to form interface feature data. Noise reduction and anti-interference processing are performed on the interface feature data to retain effective features related to the content displayed on the screen interface. The processed valid feature data is transmitted to the AI ​​unit, and the screen interface information is parsed by the interface feature recognition model to generate interface information containing operation coordinate references. The AI ​​unit generates operation guidance information based on the interface information, and realizes permissionless AI interaction through passive optical guidance or physical triggering. The touch screen device includes various touch screen devices such as smartphones, tablet computers, industrial control touch screens, vehicle touch screens, and smartwatches, and is compatible with operating systems such as HarmonyOS, Android, iOS, and Windows.

2. The method according to claim 1, characterized in that, In step 1, "no electrical connection" means: no data communication is established with the touch screen device system through any wired or wireless means such as USB, serial port, UART, OTG interface, Bluetooth, Wi-Fi; the non-visual perception module works only through external power supply or self-provided power supply, and there is no data transmission channel between it and the touch screen device system.

3. The method according to claim 1, characterized in that, The radio frequency reflection sensor operates in the frequency range of 1-10 GHz, transmits a signal power of -10 dBm to 0 dBm, and has a scanning period of 10-50 ms; the millimeter-wave scanning sensor operates in the frequency range of 24-77 GHz, transmits a signal power of 0 dBm to 5 dBm, and has a scanning period of 5-20 ms; the overlap between the signal coverage area and the touch screen display area is ≥95%; the transmission power and operating frequency configuration of the radio frequency reflection sensor and the millimeter-wave scanning sensor must ensure that the amplitude difference of the reflected signal meets the determination condition of claim 4.

4. The method according to claim 1, characterized in that, The interface feature data mentioned in step 3 includes: the difference in amplitude of reflected signals in different display areas of the screen interface, the phase offset difference between the control and the background area, and the frequency offset change of dynamic interface elements; when the amplitude difference is ≥3dB, it is determined to be different display areas, and the threshold can be dynamically adjusted to 1-2dB in special scenarios such as OLED dark mode; when the phase offset difference is ≥10°, it is determined to be the boundary between the control and the background.

5. The method according to claim 1, characterized in that, The noise reduction and anti-interference processing described in step 4 includes: using an adaptive filtering algorithm to remove environmental radio frequency interference, filtering out device housing reflection interference through signal power threshold, and using time synchronization technology to eliminate screen refresh interference; the signal-to-noise ratio of the processed signal is improved by ≥15dB compared to the original signal.

6. The method according to claim 1, characterized in that, The interface feature recognition model includes a convolutional neural network (CNN) feature extraction network and an attention mechanism interface parsing network; when the distance between the non-visual perception module and the screen is ≤20mm, the control position recognition error is ≤0.5mm; when the distance is 20-50mm, the error is ≤1.5mm; the interface content recognition accuracy is ≥95%.

7. The method according to claim 1, characterized in that, It also includes a sensing parameter adaptive adjustment mechanism: by measuring the interference intensity within the ±10MHz bandwidth of the sensor's operating frequency band through the environmental interference detection unit, when the environmental interference is ≥-20dBm, the transmission power is increased by 1-3dBm, the scanning frequency is increased to 150% of the original frequency, and the signal bandwidth is reduced to 50% of the original bandwidth.

8. The method according to claim 1, characterized in that, The non-visual perception module can be deployed in three ways: as an independent external device, as an integrated device housing, or as an external accessory integrated device. The distance between the deployment position and the screen is 5-50mm. The module's standby power consumption is ≤10μW, its operating power consumption is ≤1mW, and its size is ≤20mm×20mm×5mm.

9. The method according to claim 1, characterized in that, Supports multi-sensor collaborative operation: Radio frequency reflection sensor and millimeter wave scanning sensor can be used selectively or work simultaneously. After the feature data output by the collaborative sensors is fused, the recognition accuracy is improved by ≥5% compared with a single sensor. The implementation method of the permissionless AI interaction includes any one of physical triggering by optical signal, ultrasonic-assisted triggering, and passive optical guidance triggering.

10. The method according to claim 1, characterized in that, Adaptation for different scenarios: In automotive applications: Employs a millimeter-wave scanning sensor with a working frequency of 60-77GHz, electromagnetic interference resistance ≥20dB@60~77GHz band, scanning cycle 5-10ms, and control recognition error ≤1.0mm; Industrial control scenarios: Utilizes radio frequency reflection sensors with a working frequency of 5-10GHz, a transmission power of 0dBm~3dBm, a penetration capability of ≥80% for oil and dust, and a sensing error of ≤1mm; For small-sized devices: a miniature millimeter-wave sensor is used, with a working frequency of 24-30GHz, a module size of ≤10mm×10mm×3mm, and a control recognition error of ≤0.3mm.