A flexible, magnet-based miniaturized mechanical antenna enabling low-frequency cross-medium communication between unmanned systems Authors: Qingang Li, Zhi Cui, Xin Ma, Wei Yue, Ieng Hou U, Kangjie Zhou, Juntian Qu, Jianglei Chang, Yuping Huang, Chang Liu, Shuxiang Dong, Qinglei Hu, Yong Cui, Xining Zang Published: 2025 ABSTRACT AbstractExtremely low frequency (ELF, 3–30 Hz) signals possess strong cross-medium communication capabilities, making them particularly well-suited for underground and underwater environments. However, traditional low-frequency (LF) transmission systems are large and inefficient, posing significant limitations in practical applications. In recent studies, mechanical antennas have been explored to generate LF signals, but current approaches rely on bulky equipment with limited range, making them unsuitable for personal use or integration into small unmanned devices. To address this challenge, this study introduces a flexible, magnet-based miniaturized LF mechanical antenna, fabricated using 3D printing. The antenna consists of a macro-fiber composite layer and a flexible permanent magnet film, and features an extremely compact volume (<6.8 cm³) and low weight (<50 g). It is also highly flexible, allowing for easy integration into diverse applications. Its transmitted signal can reach 60 m before the magnetic field strength attenuates to 1 pT. Mounted on an unmanned aerial vehicle (UAV), the antenna facilitates reliable communication between quadruped robots operating outside caves and aerial robots located deep within cave interiors, where high-frequency (HF) signals cannot penetrate. This study demonstrates robust LF cross-medium communication between UAV and ground robots in cave environments, paving the way for unmanned collaboration in scenarios inaccessible to HF wireless signals.Subject terms: Electrical and electronic engineering, Mechanical engineering Introduction Reliable cross-medium communication in underground1,2 and underwater3,4 settings is essential for exploration, rescue and environmental monitoring. Electromagnetic waves at conventional radio frequencies attenuate rapidly in conductive or lossy media5,6, so high-frequency (HF) links fail after only a few meters. By contrast, low-frequency (LF) signals—whose wavelengths are orders of magnitude longer7,8—experience far less absorption and can traverse rock or seawater over practical distances9. Conventional LF transmitters employ electrically resonant antennas whose physical length must scale with the emitted wavelength (<300 kHz corresponds to >1 km)10,11. Such kilometer-scale structures are impractical for mobile or portable platforms, and because they remain electrically small, their radiation efficiency is inherently poor12,13. Mechanical antennas (MAs) sidestep the wavelength–size constraint by converting mechanical motion into time-varying magnetic or electric fields14,15. Existing MA prototypes generate LF fields by rotating or vibrating permanent magnets9,16–19, electrets20, or magnetoelectric composites21–27, and they have demonstrated longer LF links than their electrical counterparts while occupying much less volume28. However, the factors limiting the practical application of current designs are as follows: rigid permanent-magnet MAs yield strong fields but demand bulky motors and high power13, whereas lightweight electret or magnetoelectric MAs emit weaker fields insufficient for tens-of-meters links21. Here we report a flexible, magnet-based miniaturized low-frequency mechanical antenna (FMLMA) that simultaneously achieves centimeter-scale size, sub-50 g mass, and a tens‑of‑meters radiation range. The device integrates a 3D-printed flexible permanent-magnet film (FPMF) with a macro-fiber composite (MFC) actuator and is driven by a 12 V circuit weighing only 26.6 g. Operating below 10 Hz, a single FMLMA sustains detectable (1 pT) magnetic fields over 60 m while remaining conformable to curved surfaces. The proposed FMLMA antenna synergistically integrates the long-range transmission capability of permanent magnet antennas with the compactness and flexibility of electret and magnetoelectric composite designs. Its conformability allows seamless integration into space-constrained platforms, including unmanned aerial vehicles (UAVs), underwater vehicles, and other miniaturized systems, thereby facilitating the development of a three-dimensional LF communication infrastructure spanning subsurface, underwater, and open-air domains29 (Fig. 1). Critically, we demonstrate that an FMLMA-equipped UAV can maintain an LF link with a quadruped robot separated by tens of metres of rock inside a natural cave, thereby achieving cross-medium robot-to-robot communication using an MA. This capability opens a practical route for coordinated unmanned operations in environments where GPS and HF radios are ineffective. Application of the flexible magnet-based miniaturized low-frequency mechanical antenna in a three-dimensional air-land-sea communication network. By integrating the antenna into various platforms (e.g., UAVs, subterranean detectors, surface vessels, exploration ships), a 3D communication network can be established–connecting cave interiors to the outside, underwater to above-water environments, and ground to air. Results Fabrication and material characterization of FMLMA The FMLMA fabrication process involves ink preparation, printing and magnetic domain orientation (Supplementary Fig. S1). We used 3D printing to fabricate the FPMF layer on the MFC substrate, leveraging adaptability to complex geometries30,31 (Fig. 2a). By exciting the MFC layer with an electrical signal, the FPMF layer is driven to vibrate, generating an oscillating magnetic field for signal transmission (Fig. 2b). The composite film can withstand deformations exceeding 180°, facilitating conformal configuration of the FMLMA onto the UAV structure (Fig. 2c). Fabrication of the flexible magnet-based miniaturized low-frequency mechanical antenna. a In-situ fabrication of flexible permanent magnet films(FPMFs) with programmable magnetic domains on an MFC using a five-axis dispenser, and the microscopic structure of the thin film. b Schematic diagram of the flexible magnet-based miniaturized low-frequency mechanical antenna emitting magnetic field signals through mechanical vibration. c Demonstration of the antenna’s conformability and its ability to bend beyond 180°. d Remanent magnetic flux density curves of FPMFs with different magnetic powder contents (60, 70, 80, and 85 wt%). e Stress–strain curves of FPMFs with varying magnetic powder contents (60, 70, 80, and 85 wt%). f Relationship between Young’s modulus and magnetic powder content in FPMFs. g Photographs of 3D-printed FPMFs and their surface morphology. h Individual flexible permanent-magnet lines fabricated using print heads of different diameters. The FPMF primarily comprises neodymium–iron–boron (NdFeB) magnetic powder, Ecoflex, and silicone rubber32. Remanent magnetic flux density increases with NdFeB content from 60 to 85 wt% (Fig. 2d), resulting in enhanced antenna signal strength per unit volume. Young’s modulus, determined by tensile testing, rises monotonically with NdFeB loading for both magnetized and non-magnetized samples (Fig. 2e), indicating progressively reduced flexibility31. Although magnetized specimens exhibit higher absolute modulus than unmagnetized controls, the dependence of stiffness on NdFeB fraction is identical in both cases. Importantly, above 80 wt% NdFeB, further flux density gains become negligible (Fig. 2d) while Young’s modulus increases sharply (Fig. 2f), thereby reducing the antenna’s flexibility. Consequently, 80 wt% NdFeB was chosen as the optimal composition for FMLMA fabrication. The five-axis, multi-degree-of-freedom printing platform enabled the creation of FPMFs with submillimetre precision, arbitrary shapes and smooth surface finish, ensuring conformal matching to varied substrates (Fig. 2g). Flexible permanent-magnet lines printed with 0.2–1.0 mm nozzles further demonstrate the system’s resolution33,34 (Fig. 2h). Simulation-based antenna design and parameter optimization The antenna geometry (shape and dimensions) was optimized in simulation to improve performance. In a multiphysics environment (COMSOL), we constructed a model that couples electrostatics, magnetic fields, solid mechanics, and piezoelectricity. The FMLMA model comprises a piezoelectric section (including electrodes, PZT, and epoxy) and a permanent-magnet film (Fig. 3a). A voltage at a prescribed frequency was applied to excite the piezoelectric section into vibration, which in turn drove the permanent-magnet film to generate a magnetic field. The simulations investigated key aspects—antenna field strength, resonant frequency, vibration modes, magnetic-field distribution, and array gain—to guide antenna design and fabrication. Results of simulation-based antenna design and parameter optimization. a Finite-element model of the flexible magnet-based miniaturized low-frequency mechanical antenna incorporating a flexible permanent magnet film (FPMF). b Relationship between antenna field strength and remanent magnetic flux density of the flexible permanent-magnet material. c Relationship between resonant frequency and geometry. d Vibration modes and tip displacement. e 3D magnetic-field distribution during vibration. f Distance-dependent decay of radiated field strength. g Simulated figure-eight-like magnetic-field pattern. h Simulation of dual-antenna versus single-antenna strength and the dependence of array gain on element spacing. i Simulated transmission test of a flexible magnet-based miniaturized low-frequency mechanical antenna. The results show that, at fixed volume, the antenna field strength is proportional to the remanent magnetic flux density (Fig. 3b). The resonant frequency depends on both material parameters and geometry. With the width constrained by the MFC, the resonant frequency can be tuned—while keeping the total volume (stored magnetic energy) constant—by adjusting the length and thickness (Fig. 3c). Targeting 4.0 Hz, combined simulation and experimental optimization led us to select a permanent-magnet film with width 28 mm, length 45 mm, and thickness 1 mm. When operating at a frequency of 4.0 Hz, the FMLMA predominantly exhibits a first-order cantilever-bending mode, with a node–antinode displacement profile along its length; the tip displacement reaches 18.3 mm (Fig. 3d). The surrounding magnetic field is dipole-like (Fig. 3e, g); its magnitude decays approximately with the inverse cube of distance, consistent with theory (Fig. 3f). To further enhance field strength, array summation was evaluated. Simulations indicate that a two-element array arranged in the same direction provides a twofold gain over a single element, and this gain remains essentially invariant as the inter-element spacing varies from 5 to 100 cm (Fig. 3h). By controlling the drive voltage, amplitude-shift keying(ASK) was implemented in simulation to transmit the message “Hello”, verifying the feasibility of this scheme for signal transmission (Fig. 3i). Signal transmission performance of the FMLMA To examine how antenna characteristics influence transmission performance, we fabricated two antennas optimized for different operational frequencies: a 3.1 g antenna operating at 4.0 Hz and a 48.2 g antenna operating at 1.7 Hz. The oscillation patterns of the two antennas were captured by a high-speed camera to observe their motion and stability during vibration (Fig. 4a), and a laser distance meter was used to measure the amplitudes of both antennas. The antenna achieved maximum amplitude at its resonant frequency (Fig. 4b). Recognizing that greater vibration amplitude leads to longer transmission distances16, we focused experimental validation on amplitude characterization. The comparative results of the two antennas show that lowering the resonant frequency enhances the FMLMA’s amplitude (Fig. 4a), but may compromise efficiency and increase antenna size. Therefore, frequency selection must be tailored to specific scenarios. Fine-tuning the antenna’s geometry allows it to operate within a 1–10 Hz range. Performance testing of the flexible magnet-based miniaturized low-frequency mechanical antenna. a Vibration images of mechanical antennas with resonant frequencies of 1.7 Hz and 4.0 Hz, captured by a high-speed camera at 480 frames per second. b Relationship between the excitation frequency and vibration amplitude of the mechanical antenna. c 3D angular variations of Bx, By, and Bz; the coordinate system is right-handed with +x from antenna to fluxgate, +y vertically downward, and +z completing the triad (normalized). d Raw image of the mechanical antenna’s radiated signal and the signal extracted via Fourier transform. e Frequency dependence of signal strength measured at different distances for the flexible magnet-based miniaturized low-frequency mechanical antenna with resonant frequencies of 1.7 Hz and 4.0 Hz. f Inverse cubic relationship curve between signal strength and distance. g Theoretical and experimental signal strength curves of a dual-antenna array (normalized). The directional characteristics help focus the radiated energy, thereby extending the transmission range. The FMLMA’s radiation pattern is similar to a dipole’s: in the plane perpendicular to the vibration direction, field strength is maximal along the antenna’s length and minimal in the transverse direction (Fig. 4c). In subsequent tests, we aligned the antenna’s strongest gain direction with the transmission path. As the resonant frequency is fixed by the antenna’s mechanical structure, communication is implemented through binary encoding by starting and stopping transmissions. Received signals are processed using Fourier transform-based filtering to extract and decode information at the designated frequency (Fig. 4d). This will filter out noise from other frequency bands (e.g., power frequency signals) and retain only the target signal. The communication range is a critical performance indicator for the FMLMA. We first measured how the radiated intensity of the antenna varies with distance. The magnetic fields detected by the fluxgate sensor at different positions of the FMLMA (Fig. 4e) indicate that the radiation intensity is inversely proportional to the cube of the distance (Fig. 4f). The curve of FMLMA radiated magnetic field strength versus the inverse of the third power of distance reveals a strong linear relationship, with an R² value exceeding 0.9999 (Fig. 4f). Based on this trend, the FMLMA’s signal is expected to remain above 1 pT out to approximately 60 m (the minimum detectable magnetic field intensity of current high-precision magnetic field sensors, such as SQUID, is approximately 1 pT). By employing a two-antenna array configuration, and taking advantage of the low operating frequency of FMLMA signals—which simplifies phase synchronization and minimizes alignment errors18,35—we achieved a 1.74‑fold increase in magnetic field strength compared to a single antenna (Fig. 4g). The two antennas in the array are arranged in parallel and aligned in the same direction, with a 15 cm spacing, and driven by the same excitation source to maintain consistent initial phase. Specifically, compared to HF antennas, LF antennas benefit from two main advantages in phase synchronization: first, the significantly longer wavelength relative to the antenna spacing greatly reduces phase errors caused by geometric positioning deviations; second, the longer oscillation period effectively mitigates the relative impact of asynchronous startup times between antennas, thereby minimizing initial phase discrepancies. Notably, this work presents an initial exploration of the antenna array’s potential, in experiments, manufacturing tolerances led to small mismatches in resonant frequency and other errors, and the ideal twofold gain was therefore not fully achieved. Further improvements—such as precisely controlling antenna dimension tolerances to ensure identical resonance frequencies, and introducing closed-loop timing control to enhance phase synchronization—could enable the magnetic field strength of the antenna array to approach nearly twice that of a single antenna. Durability and stability assessment of the FMLMA We tested long-term durability and stability of the FMLMA before and after subjecting it to 1000 bending–unfolding cycles beyond 90° to assess whether the antenna maintains long-term stable performance after conformal integration onto equipment. The signal processing procedure for the original measured data (Fig. 5a) involved acquiring magnetic field signals with a fluxgate sensor, followed by filtering to isolate the 4.0 Hz frequency (Fig. 5b, c) and applying short-time Fourier transform (STFT) to extract amplitude variations (Fig. 5d). The antenna’s amplitude remained stable within a 1% fluctuation range, with no discernible trend over time. Fatigue testing of the flexible magnet-based miniaturized low-frequency mechanical antenna and the effect of bending on performance. a Raw magnetic field data from the fatigue test. b Sampled magnetic field data from the fatigue test. c Processed magnetic field data after applying a 4.0 Hz filter. d Variation in the mechanical antenna’s vibration amplitude over time. e One thousand bending–unfolding cycles beyond 90°. f Raw magnetic field data from the second fatigue test. g Sampled magnetic field data from the second fatigue test. h Processed magnetic field data from the second fatigue test after applying a 4.0 Hz filter. i Variation of the mechanical antenna’s vibration amplitude over time during the second fatigue test. To assess conformability and mechanical durability, the FMLMA was subjected to 1000 bending–unfolding cycles beyond 90° (Fig. 5e), followed by a fatigue test. The second test data (Fig. 5f) were similarly filtered (Fig. 5g, h), and STFT analysis revealed minimal amplitude variation (within 1%) (Fig. 5i). The mean post-bending amplitude deviated by less than 0.5%, demonstrating the FMLMA’s exceptional durability and signal stability. We measured the magnetic field strength at five points 1 mm from the surface of the antenna after bending and took the average, which was 62.1 mT; before bending, the magnetic field strength at the same location was 62.2 mT, showing no significant attenuation. Cross-medium communication capability of the FMLMA We conducted a through-the-earth (TTE) transmission test at the Lee Shau Kee Science and Technology Building. A “Hello” voice signal was transmitted from a basement approximately 20 m underground to the earth’s surface, with signal strengths measured at each floor (Fig. 6a). The detailed procedure of signal encoding, transmission, and decoding is described in Methods and Supplementary Video 1. Cross-medium transmission attenuation and transmission rate testing. a Schematic diagram of the 20-m underground communication test using the flexible magnet-based miniaturized low-frequency mechanical antenna(FMLMA). b Curve of signal strength versus penetration distance in the through-ground experiment. c Signal attenuation curve of the FMLMA when transmitting and receiving signals within the same mine tunnel (theoretical vs. experimental values) (normalized). d Signal attenuation curve of the FMLMA when transmitting through rock layers in the cave (theoretical vs. experimental values) (normalized). e Schematic diagram of the FMLMA water-penetration experiment. f Signal attenuation curve of the FMLMA in the water-penetration experiment (theoretical vs. experimental results) (normalized). g Relationship between the cross-medium attenuation coefficient and propagation distance at different frequencies. h Signal-to-noise ratio and bit error rate of the antenna operating at 1.7 Hz at different transmission rates. The top waveform plot shows the antenna transmitting “Hello” at 0.05 bit/s, while the two lower curves show the threshold-decoding results using (1) the full waveform of each bit and (2) only the last 15% of each bit’s waveform. i Signal-to-noise ratio and bit error rate of the antenna operating at 4.0 Hz at different transmission rates. The top waveform plot shows the antenna transmitting a “Hello” at 0.5 bit/s. The three waveform segments below are instances of the letter “l” appearing in different contexts (the two “l”s in “Hello” and another “l” in a different word), all with Pearson correlation coefficients above 0.99. Due to significant attenuation by rock layers and concrete structures, conventional HF signals (e.g., mobile phone and GPS signals) were completely blocked in the basement. In contrast, the FMLMA can radiate signal strengths consistently above 20 pT when transmitted to the surface. Inset of Fig. 6b presents the magnetic field strength at various locations in the frequency domain, while the magnetic field vs. distance curve indicates that during cross-medium transmission from underground to the surface, the inverse-cube relationship between radiation strength and distance holds (Fig. 6b). Quality factor Q of the FMLMA is quantified to characterize the near-field communication efficiency. According to the bandwidth method, the quality factor is calculated using the formula Q=f0BW, where f0 represents the resonant frequency, and BW represents the bandwidth. The bandwidth can be determined by measuring the frequency interval at which the amplitude decreases to 22 (approximately 0.707) of its peak value on a linear scale. The Q-factor of the FMLMA exceeded 105. A high Q-factor of FMLMA indicates lower energy losses, greater emitted field strength at the same power, and precise energy accumulation at a specific frequency over multiple cycles, which is beneficial for long-range signal transmission. Moreover, the prolonged ring-down tail inherent to a high Q-factor can be effectively mitigated by employing optimized modulation schemes25. We conducted field tests in an underground mining cave to demonstrate FMLMA communication within a single tunnel (primarily through air, Fig. 6c) and between separate tunnels through intervening rock (Fig. 6d). Signal strength at identical distances was 37.2 nT in air and 36.7 nT in rock, indicating only a 1.34% attenuation through rock—close to the theoretical attenuation36 (<1%) (see Supplementary Section S1). In contrast, conventional HF signals attenuated almost immediately under these conditions: a mobile phone lost signal entirely, and a walkie-talkie (operating around 410 MHz) managed a range of under 20 m through rock. To validate the FMLMA’s underwater transmission capability, we conducted experiments in a seawater pool (Fig. 6e). The LF signal strength in seawater was about 12% lower than at the same distances in air (Fig. 6f). For HF signals, seawater’s much higher conductivity and relative permittivity compared to rock leads to significantly stronger reflection and refraction of electromagnetic waves. At the LF used here, however, the ≈12% discrepancy arises primarily from wave-induced tilts of the pool surface and local surface curvature, which misalign and spread the near-field magnetic signal relative to the fluxgate sensor (see Supplementary Section S2). Extrapolation of the trend indicates the signal can still travel about 52.4 m before the magnetic field strength drops below 1 pT. This remains a significant advantage over HF communication methods like walkie-talkies (which have ≈3 m range in seawater) and mobile phones (≈30 cm in seawater). Theoretical results show that HF signals attenuate much more rapidly than a signal at the FMLMA’s operating frequency, both in the ground and in seawater (Fig. 6g). This explains why a walkie-talkie that can transmit over several kilometers in air has a communication range of no more than 20 m in a cave environment, and only about 3 m in seawater, with mobile phone signals performing even worse. Antenna communication primarily involves three processes: encoding, transmission, and decoding. The encoding is completed by a chip in the antenna driver, with the encoding time for five characters being less than 20 ms; decoding is performed by the host at the magnetic field sensor end, with the decoding time for five characters being less than 10 ms. Therefore, the main factor limiting transmission efficiency is the speed of information transmission. We tested the transmission rates of two antennas operating at 1.7 Hz and 4.0 Hz. The five letters of “Hello” were each converted into 8-bit binary (0/1) code, totaling 40 bits, as the message for transmission. On the receiving end, the processor extracted the information at the corresponding frequency from the magnetic field signal and used a threshold method to determine 0/1 bits. The signal-to-noise ratio (SNR) (the strength ratio of ‘1’ vs ‘0’ signals) and the bit error rate (BER) (the number of failed bits divided by the total number of bits transmitted) were used to characterize the accuracy and reliability of signal transmission. For the antenna operating at 1.7 Hz, we tested eight transmission rates from 0.05 to 1.00 bit/s (Supplementary Fig. S2a–h). As the transmission rate decreased, the SNR gradually increased. At 0.10 bit/s, the SNR reached 4.58 and the bit error rate dropped to 0 for the first time, meaning the signal could be easily identified with a simple threshold method. Because the 1.7 Hz antenna has a lower frequency and larger amplitude, its settling and ring-down times are longer than those of the 4.0 Hz antenna, resulting in a ginkgo-leaf-shaped waveform in the signal plot (Fig. 6h). Considering this characteristic, we used only the last 15% of each bit’s waveform for decoding, and found that the SNR increased significantly and the bit error rate decreased accordingly (Fig. 6h). For the antenna operating at 4.0 Hz, we tested nine transmission rates from 0.33 to 8.00 bit/s, and its SNR and BER showed the same trends as the antenna operating at 1.7 Hz (Supplementary Fig. S2i–q). At 2.00 bit/s, the SNR reached 6.19 and the bit error rate dropped to 0 for the first time (Fig. 6i). Because the 4.0 Hz signal has a shorter relaxation time, it supports a higher data transmission rate compared to the 1.7 Hz signal, which is reflected in the waveform by a more distinct separation between “0” and “1”. In our experiments, although transmission rates above 4 bit/s led to bit errors up to 17.5%, the waveforms for the same letter (for example, the two “l”s in “Hello” and another “l” appearing in a different word) were highly consistent (Fig. 6i and Supplementary Fig. S2j). The Pearson correlation between “l” waveforms at different positions exceeded 0.99, whereas the correlation between two randomly selected 1-bit waveforms was between 0.52 and 0.89 (Supplementary Fig. S2r). Based on this finding, we succeeded in decoding the signal by recognizing letters using the correlation coefficient method at 4 bit/s. UAV–quadruped robot LF communication demonstration in caves Moreover, the long signal transmission distance and cross-medium capability of FMLMA enables collaborative communication among unmanned agents in environments that obstruct HF signals, where radio frequency, GPS, and vision-based communication technologies fail. Here we developed a UAV-mounted FMLMA signal transmission system, in collaboration with a quadruped robot, designed for cave exploration and rescue (Supplementary Fig. S3). The FMLMA communication system enables a simultaneous advantage of the UAV’s broad scanning range37,38 and the quadruped robot’s robust ground capabilities39,40 in cave environments where the HF signals are obstructed (Fig. 7a). We evaluated via simulation the effect of UAV-induced motion (dominated by 1–10 Hz LF disturbances) after integrating the FMLMA onto the UAV. The results show that motion-induced noise leads to a noticeable near-range reduction in magnetic-field amplitude, with the SNR at 1 m decreasing to approximately 42; nevertheless, because the nominal near-range SNR is high, such disturbances do not materially impair link reliability. At longer ranges, the influence of motion decays rapidly, and background noise remains the dominant factor at the range limit. For antenna arrays, the previous simulation (Fig. 3h) demonstrated that the variation in array element spacing caused by UAV-induced motion does not affect the array gain. Subsequent field tests that successfully transmitted coordinates over longer distances corroborated these findings. Unmanned Aerial Vehicle(UAV) and quadruped robot collaborating via Low-Frequency communication in a cave. a Schematic diagram of the communication link between the UAV and the quadruped robot. b Image of the test cave, with red flags marking designated points. c UAV detecting the target location inside the cave. d UAV uses the onboard flexible magnet-based miniaturized low-frequency mechanical antenna to transmit signals. e Signal processing workflow for UAV–quadruped robot collaborative communication. f Determining binary values using an amplitude threshold on the short-time Fourier transform processed signal. g Processed coordinate information represented in binary signal format. h The quadruped robot waiting outside the cave for information transmitted by the UAV. i Based on the coordinates sent by the UAV, the quadruped robot enters the cave and reunited with the UAV. j The UAV flew out of the cave, while the quadruped robot remained inside to carry out subsequent work. Considering temperature variations encountered in caves and underwater environments, we characterized FMLMA performance at 2 °C and 40 °C relative to room temperature (25 °C). The field strength exhibits slight attenuation at both high and low temperatures, attributable to temperature-induced changes in the Young’s modulus of the permanent-magnet film, which shift the resonant frequency and reduce vibration amplitude. The effect is modest: the transmission distance decreases by about 6.1% at 2 °C and 1.3% at 40 °C, which does not compromise system usability. We simulated a realistic mine rescue scenario with rugged, water-crossing terrain and other obstacles hindering human access (Supplementary Video 2). Inside the cave, five red flags were placed to mark target points (Fig. 7b), and we established a common origin and coordinate system for both the UAV and the quadruped robot. Once the UAV detected a red flag, the corresponding coordinates were encoded in binary and transmitted to the standby quadruped robot outside the cave (Fig. 7c, d). After the UAV’s signal was received by the fluxgate sensor, the carrier frequency component was extracted and its amplitude was analysed using STFT (Fig. 7e). A threshold method then differentiated the ‘0’ and ‘1’ based on the signal amplitude (Fig. 7f), decoding the coordinate information (Fig. 7g, h). The quadruped robot reunited with the UAV within one minute according to the as-received-and-decoded coordinates (2, 3, 1) in this case (Fig. 7i); if further exploration was required, the above process was repeated, and once the exploration was complete, the UAV departed and the quadruped robot took over the ground rescue task (Fig. 7j). Effective real-time communication in cave environments will enable a broadly applicable LF communication strategy for unmanned intelligent agents’ navigation and collaboration in caves and other complex environments. Furthermore, we can convert the signals from LF at the water-air or ground-air interface to HF, which relays the transmission to further communication distance. For example, the UAV could use LF signals to send information to a transmission tower, which are then converted to HF and relayed to a remote-control station (Fig. 7a). Discussion and conclusion This study proposes a flexible MA based on FPMF and MFC to overcome the large size and low efficiency of traditional LF antennas, as well as to resolve the trade-offs of balancing stiffness, size, and transmission distance in similar MAs. To optimize both flexibility and transmission distance—key factors—an optimal composition of NdFeB powder was selected in the range of 60–85 wt%. The use of 3D printing for fabricating the FPMF allows for adaptation to various surfaces, thereby broadening its potential applications. The antenna’s resonant frequency was tuned through structural design to the range of 1–10 Hz. Its radiation pattern resembles that of a dipole antenna, following which the transmission range can be extended by directionally concentrating the radiation energy. The FMLMA is capable of achieving communication over a distance of 60 m, within the detection precision of the current sensors. Additionally, a two-antenna array yielded a 1.74-fold increase in signal strength by phase synchronization, indicating that the array summation method can further extend the effective communication distance. The FMLMA shows significant advantage of low radiation attenuation in rock (1.5% lower than in air), and in seawater (12% lower than in air). The antenna’s transmission rate can reach 2 bit/s (using threshold method to distinguish 0/1). By analyzing the signal correlation at higher transmission rates, we demonstrated that the optimization algorithm enables effective communication at transmission rates of 4 bit/s and beyond. While the present data rate is relatively limited (2–4 bit/s), this mainly stems from the inherently narrow bandwidth of LF antennas. In underground, cave, and underwater high-loss environments, however, such data rates are sufficient for transmitting key commands to coordinate unmanned systems in high-loss environments, though they may be limited for more data-intensive applications. Future improvements may be achieved through optimized modulation schemes and hybrid communication strategies. Compared to similar designs (Table 1), the MA developed in this study offers significant advantages. It features a compact form factor and long transmission range, along with flexibility not available in comparable systems. Table 1Comparison of various LF MA designs from literatureTypes of MAsFrequencyaMaterialFlexibilityMagnetic field intensity (measured at distance)Antenna SizeExternal DriveRadiation Intensity at 1 mbRefs.Permanent Magnet-Based MA15.48 HzN52 NdFeB & PZTRigid280 nT (1 m)7 cm3Commercial voltage amplifier280 nT1628/32 HzNdFeBRigid50 nT (10 m)189 cm3Motor50 μT1775-125 HzNdFeBRigid59.3 nT (2 m)54 cm3Motor474 nT18715 HzNdFeBRigid0.7 fT (1 km)105 cm3Motor700 nT35Piezoelectric Polymer-Based MA20 HzFEP/THVRigid1 fT (136.1 m)500 cm3Motor19 pT20Magneto-Electric Composite-Based MA35.5 kHzLiNbO3Rigid–(Effective transmission distance: 30 m)18.9 cm3No External Drive–2580 HzCu & TENG SystemRigid–(Effective transmission distance in a tube: 100 m)-Control the volume of the medium.-2637.95 kHzPZT & Terfenol-DRigid15 nT (0.3 m)8.8 cm3Uniform Magnetic Fieldc406 pT2322.23 kHzPZT & MetglasRigid6 pT (5.5 m)35 mm3Uniform Magnetic Fieldc998 pT21This Work1–10 HzMFC & FPMFFlexible1 pT (60 m)10 cm3 (Antenna system)No External Drive215 nTaIn some references, the frequency range is not explicitly stated, so the frequencies they used in paper are listed here.bPartially based on calculated values.cAn external magnetic field is required for the polarization of the magnetostrictive layer. Comparison of various LF MA designs from literature aIn some references, the frequency range is not explicitly stated, so the frequencies they used in paper are listed here. bPartially based on calculated values. cAn external magnetic field is required for the polarization of the magnetostrictive layer. Furthermore, we developed FMLMA based signal transmission systems with low weight down to 30 g (operating at 4.0 Hz) and 80 g (operating at 1.7 Hz), which are conformally integrated onto a UAV enabling its communication with a quadruped robot in HF-obstructed environments. The system facilitated efficient communication across complex terrains, including rugged and waterlogged areas, enabling autonomous unmanned agents’ collaborations. Methods FMLMA fabrication method The core process for fabricating the FMLMA is the production of the FPMF, which involves three main steps: ink preparation, 3D printing, and magnetic domain orientation. Ink preparation We selected Ecoflex 00-30 Part B (Smooth-On Inc.), SE 1700 (Dow Corning Corp), and NdFeB powder (5 μm particle size, MQFP-B-2007609-089, Magnequench) as the primary components of the FPMF. Ecoflex 00-30 Part B serves as the flexible substrate, SE 1700 Base as a facilitator for hardening, and NdFeB powder as the magnetic component. In this formulation, the magnetic powder content was adjusted between 60–85 wt%, with the proportions of the other components adjusted accordingly. The final composition is as follows: 80.0 wt% NdFeB powder, 6.67 wt% SE 1700 Base, 0.67 wt% SE 1700 catalyst, and 12.66 wt% Ecoflex 00-30 Part B. After the materials were mixed in the specified proportions, the ink was thoroughly stirred at 200 rpm using an overhead stirrer. Finally, a five-minute vacuum degassing process was performed. 3D printing We used a five-axis dispensing machine to perform additive manufacturing of the FPMF on the MFC surface. The prepared ink was loaded into the extrusion mechanism and printed onto the MFC substrate in situ according to a predefined pattern. The printing parameters were as follows: the printhead movement speed was 8 mm/s; the extrusion pressure of the slurry was 30 MPa, with a nozzle diameter of 0.8 mm, and a printing volume of approximately 5 mm³ per second; the delay time for the adhesive opening was 0.02 s. To ensure uniform and stable internal magnetic fields, the fill density was set to 100%, and the pattern path was planned using a linear fill method. Upon completion of printing, the composite structure comprising the FPMF and MFC layers was placed on a heating platform at 120 °C for 90 min. After curing, the FPMF layer was tightly bonded to the MFC layer without any additional processing. Magnetic domain orientation The cured FPMF–MFC composite was placed into the groove of a magnetizer (model SYF-2070) equipped with external coils to generate the required magnetic field. The magnetization voltage was adjustable from DC 1000 V to 2000 V, the capacitance was 7000 μF, and the maximum instantaneous output current was 10 kA. Upon activation, a pulsed magnetic field of approximately 3.0 T was generated within the groove, orienting the magnetic domains within the FPMF. Finite-element simulation methodology for the FMLMA An FMLMA finite-element model was established in COMSOL. The geometry consists of an MFC and a FPMF. Following the structure of model M8528-P1, we built the MFC geometry with materials including PZT, epoxy, and copper electrodes. The FPMF was represented by an equivalent thin rectangular layer; its material was user-defined using experimentally measured parameters such as remanent magnetic flux density, density, and Young’s modulus. The physics included Electrostatics, Solid Mechanics, and Magnetic Fields, with the Piezoelectric Effect enabled for multiphysics coupling. A drive voltage was applied to the copper electrodes in the Electrostatics module; one end of the MFC was fixed in the Solid Mechanics module; and the magnetization polarity of the flexible magnet was specified in the Magnetic Fields module. To study the relationship between the FMLMA resonant frequency and the distribution of the flexible magnet on the MFC surface, we performed a parametric sweep over the magnet layout and carried out resonant frequency studies to compute the characteristic frequencies under different layouts. For the spatial magnetic-field distribution, a spherical air domain was added and a Frequency-Domain study was conducted, from which the magnetic field in the surrounding air domain was obtained. Mechanical and magnetic performances testing of FMLMA. Mechanical property testing of the FPMF We evaluated FPMFs with varying NdFeB contents (60, 70, 80, and 85 wt%) to determine their mechanical property. The test procedure was carried out in accordance with GB/T 528-2009 (Supplementary Fig. S6 and Supplementary Video 3). Considering that changes in internal magnetic domain distribution after magnetization might affect the mechanical properties of the material, comparative experiments were conducted between magnetized and non-magnetized samples. Magnetic radiation performances testing of FMLMA We first measured how the radiated intensity of the antenna varies with distance. The testing system is primarily composed of an FMLMA, a control module, an excitation module, a fluxgate sensor (MAG-13MSL, Bartington), a digital acquisition card, and a processing terminal (Supplementary Fig. S7). By fixing the transmitting antenna in place and moving a fluxgate sensor to different distances, we recorded the signal strength at each point. The fluxgate sensor sampled at 5 kHz, and at each location data were collected for at least 20 s to ensure stable frequency-domain amplitude measurements. The signal processing procedure for the original measured magnetic field data was as follows: magnetic field signals were initially acquired using a fluxgate sensor. A filtering process was then applied to isolate the amplitude of the target frequency signal. Subsequently, an STFT was performed to characterize the amplitude–time relationship, thereby extracting how the signal amplitude varies over time. Drive circuit and control method The drive circuit consists of a signal control module and a voltage amplification module (Supplementary Fig. S8), which together generate the necessary voltages to drive the FMLMA at its resonant frequency. The signal control module has a 12 V input voltage and can output a frequency-controllable 0–5 V sine wave. By modulating the control module’s output voltage, we achieved FMLMA communication using an ASK scheme. The voltage amplification module is a commercial product (Harbin Xinmingtian Technology Co., China; model AMD-HV2000) with a 12 V input and an input signal range of 0–5 V. This module has two output terminals: one outputs a fixed 500 V, while the other outputs a voltage range of 0–2000 V depending on the input signal. By integrating the signal control chip (Supplementary Fig. S9) and the voltage amplification chip, the total mass of the driving circuit is kept below 26.6 g, making it suitable for mounting on most small UAVs and underwater vehicles. Voice signal transmission method To illustrate the information processing and transmission mechanism of FMLMA cross-medium communication, we conducted an underground-to-surface transmission test using the voice signal “Hello” as an example. The specific process was as follows: the microphone received voice signals and transmitted them to the processor, where they were converted into textual information. During the encoding process, textual information was converted into corresponding ASCII codes and subsequently into binary signals. These binary signals were then sent to the FMLMA’s signal control module, which output voltages corresponding to the binary input (high for ‘1’, low for ‘0’). The voltage signals were amplified by an excitation module, driving the FMLMA to vibrate and generate magnetic field signals. When the fluxgate sensor received these magnetic field signals, the receiving-end processor extracted specific frequency components and then converted the magnetic field signals back into binary signals according to their amplitude. Subsequently, these binary signals were sequentially decoded (the reverse of the encoding process) into ASCII codes and textual information. Finally, the processor converted the text back into an audio signal, which was played through a speaker (see Supplementary Video 1 for a demonstration). Through-rock tunnel communication test Through-rock tunnel tests were carried out in the Yanji Tianbaoshan Mining Cave (Jilin Province, China) to compare FMLMA performance against conventional HF communication methods (mobile phone and walkie-talkie). To assess signal attenuation in air versus rock under identical conditions, we conducted two complementary experiments. In the first, the antenna and fluxgate sensor were co-located within a single straight tunnel, so that the magnetic field propagated solely through air. In the second, the antenna and sensor were situated in two adjacent tunnels, requiring the signal to penetrate the intervening rock. Communication tests were also performed under identical conditions using mobile phones and walkie-talkies as HF controls. Distances in the straight-tunnel configuration were measured directly with a laser rangefinder; for the dual-tunnel case, we used a tape measure and protractor to determine the folded path length and angles between segments, from which the straight-line rock-penetration distance was calculated. During each test, the antenna remained fixed while the fluxgate sensor was moved to successive measurement points to record local magnetic field strength. The sensor sampled at 5 kHz, with each position logged for at least 20 s to ensure frequency-domain amplitude stability. Raw time-series data were Fourier-transformed to extract and record the amplitude of the 4.0 Hz magnetic component. Through-seawater pool communication test Through-seawater pool tests were performed in a seaside saltwater pool (1.35 m deep; ~50 m² area) in Weihai, Shandong Province, China, where the seawater’s conductivity was approximately 4.0 S/m and the temperature was 25 °C. Water depth was verified using a sounding rod. The FMLMA was mounted directly above the water surface, and the fluxgate sensor was immersed at depths of 1.0, 1.1, 1.2, and 1.3 m by a fixed mounting. At each depth, the sensor sampled at 5 kHz for a minimum of 20 s. The recorded time-domain signals were Fourier-transformed, and the amplitude of the 4.0 Hz component was extracted and logged. Extreme-temperature experimental protocol Using the 40 °C test as an example: the antenna, fixture, PCB, and other ancillary components were placed in an oven set to 40 °C. After the thermometer stabilized (≈1 h), power was applied and magnetic-field strength measurements were taken. The 2 °C and 25 °C tests followed the same procedure, using a refrigerator and a room-temperature environment, respectively. LF signal transmission system based on UAV and quadruped robot An integrated experiment was performed in a natural cave near Tianjin, China, to demonstrate cross-medium communication between a UAV and a quadruped robot via an LF communication link. The UAV platform (DJI Matrice 30) carried the FMLMA, its drive electronics, and a custom 3D-printed resin support structure that secured both the antenna and the driver. During cave flights, the UAV used onboard vision and infrared sensors to adjust and stabilize its attitude. The UAV’s onboard 24 V DC supply was stepped down to 12 V via a voltage-conversion module to power the FMLMA. On the ground, the quadruped robot (Unitree Go2) was connected to the fluxgate sensor’s processing terminal to receive the decoded coordinates and navigate to the designated location. Supplementary information Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Video_1 Supplementary Video_2 Supplementary Video_3 Acknowledgements This work was supported by the National Natural Science Foundation of China (Grant Nos. 52105459, 62271029) and the Beijing Nova Program (Grant No. 20220484056). Author contributions Q.L. carried out the majority of the experiments and drafted the manuscript. Z.C. completed the main simulation. X.M., I.H.U., K.Z., J.Q., J.C., Y.H., C.L., and S.D. assisted with conducting the experiments. W.Y. assisted with data analysis and contributed to manuscript writing. Q.H., Y.C., and X.Z. supervised the project, secured funding, provided resources, and reviewed and edited the manuscript; X.Z., as lead corresponding author, additionally coordinated project administration and handled all correspondence with the journal. Peer review Peer review information Communications Engineering thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: [Wenjie Wang, Rosamund Daw]. A peer review file is available. Data availability The datasets generated and analysed during this study have been deposited in Figshare (DOI: 10.6084/m9.figshare.30415189). Further information and requests for resources or data should be directed to the corresponding author. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Qinglei Hu, Email: huql_buaa@buaa.edu.cn. Yong Cui, Email: cuiyong@buaa.edu.cn. Xining Zang, Email: xzang@tsinghua.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s44172-025-00569-1. References Associated Data Supplementary Materials Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Video_1 Supplementary Video_2 Supplementary Video_3 Data Availability Statement The datasets generated and analysed during this study have been deposited in Figshare (DOI: 10.6084/m9.figshare.30415189). Further information and requests for resources or data should be directed to the corresponding author.