Abstract

Millimetre-wave (mmWave) radars are key enabler in the upcoming 6G era for high-resolution sensing and detection. Conventional photonic radars are mostly realized in tabletop systems composed of bulky discrete components, whereas the more compact integrated photonic radars are difficult to reach the mmWave bands owing to the unsatisfactory bandwidths and signal integrity of the underlying electro-optic modulators. Here we overcome these challenges and demonstrate a centimetre-resolution compact photonic mmWave radar based on a 4-inch wafer-scale thin-film lithium niobate (TFLN) technology. The TFLN photonic chip consists of a first electro-optic modulator for generating broadband radar waveforms via optical frequency multiplication, and a second modulator for de-chirping the received echoes. This greatly relieves the bandwidth requirements for the digital-to-analogue converter in the transmitter and analogue-to-digital converter in the receiver. Operating in the mmWave V band (40–50 GHz), we achieve multi-target ranging with a resolution of 1.50 cm and velocity measurement with a resolution of 0.067 m s⁻¹. Furthermore, we construct an inverse synthetic aperture radar with a two-dimensional resolution of 1.50 cm × 1.06 cm. Our integrated TFLN photonic mmWave radar chip provides a compact and cost-effective solution in the 6G era for high-resolution sensing and detection in vehicle radar, airborne radar and smart homes.

Main

For decades, radio detection and ranging (radar) at microwave frequencies has been the fundamental technology for various applications, such as airborne object detection, weather forecasting, resource exploration and vital-sign monitoring¹. In the forthcoming 6G era, millimetre-wave (mmWave) radars operating at even higher frequencies above 30 GHz and with broader bandwidths are anticipated to play a pivotal role in integrated sensing and communication systems that require high-resolution detection and real-time situational awareness, enabling new application scenarios such as indoor sensing, automated driving and vital-sign monitoring^2,3. However, the operation frequency and bandwidth of traditional electrical radar systems are typically limited and often traded off with each other, making it challenging to precisely locate, recognize and image objects with large detection ranges and fine resolution simultaneously. While advanced technical solutions, such as phased array technology, multiple-input multiple-output architecture, super-resolution algorithms and compressed sensing, can be applied to achieve precise detection, they often come at the cost of substantially increased system complexity⁴.

The emergence of photonics-based radar technology provides a promising solution to overcome these limitations by processing microwave signals in the optical domain. This leverages the benefits of photonics systems, including high frequency, large bandwidth, low transmission loss, reconfigurability and anti-electromagnetic interference^5,6,7. In 2014, the first photonic-assisted fully digital radar (PHODIR) was demonstrated for the detection of a non-cooperative aircraft⁸. The reported ranging resolution of 23 m, however, was limited by the narrow signal bandwidth of 200 MHz. To further improve the resolution, different architectures of microwave photonic radars have been proposed, such as frequency multiplication^{9,10,11,12,13,14,15}, optical injection of semiconductor lasers¹⁶, photonic stretch processing¹⁷, cyclic frequency shift¹⁸, dual optical frequency comb-based frameworks with signal stitching¹⁹ and frequency up-conversion^20,21,22. The detailed performances of these methods are shown in Extended Data Table 1. However, most microwave photonic radars so far are still constructed using discrete optoelectronic devices with substantial disadvantages in terms of size, weight and power consumption (SWaP).

Recently, integrated photonics has opened up new opportunities for improving the SWaP performance of microwave photonic systems by miniaturizing and integrating multiple photonic devices in chip-scale systems^{23,24,25,26,27}. Benefiting from the compatibility with mature complementary metal–oxide–semiconductor (CMOS) fabrication technology, several silicon (Si) photonic radars have been realized in the microwave bands^28,29,30, with a maximum demonstrated bandwidth of 6 GHz (from 12 GHz to 18 GHz) (Extended Data Table 1). However, the modulation mechanism used by silicon electro-optic modulators, that is, free carrier depletion, presents inherent limitations to the modulation bandwidths, linearity and extinction ratios, substantially impacting the operational frequency range of the radars and the signal quality (for example, spurious harmonics) of their waveforms. As a result, although integrated Si photonic radars have achieved a relatively high level of integration, including modulators and photodetectors³⁰, they have not been able to reach the mmWave bands (>30 GHz), which are highly desired in future indoor sensing, autonomous driving, as well as 6G-based imaging and sensing networks^2,5.

Thin-film lithium niobate (TFLN) platform is an excellent candidate to address these challenges and bring the operation frequency of integrated photonic radars to the mmWave bands. On the one hand, TFLN exhibits a fast and linear Pockels effect, making it well suited for achieving high-speed and linear electro-optic modulators (EOMs). On the other hand, the tight optical confinement in TFLN devices enables the implementation of multiple photonic functionalities in a single TFLN photonic integrated circuit^31,32,33. In recent years, a number of TFLN-based EOMs have been developed, achieving unprecedented performance metrics, including ultrabroad modulation bandwidths^34,35,36,37 and CMOS-compatible drive voltages³⁸. Efforts to further integrate these components into chip-scale systems through wafer-scale fabrication have led to integrated microwave photonic system signal processors with unparalleled speed and power-consumption performances³⁹. These collective achievements have paved the path for the TFLN platform to be applied to photonic mmWave radar applications that require high frequency, large bandwidth and compact form factor at the same time.

In this Article, we demonstrate a compact photonic radar system operating in the mmWave V band based on a TFLN photonic circuit, achieving centimetre-level resolutions. Fabricated from a 4-inch wafer-scale process, the TFLN photonic radar chip consists of a frequency multiplying module for mmWave radar waveform generation and a frequency de-chirp module for echo signal reception. Benefiting from the broad bandwidths of all photonic components in our chip and the no-filter design, the centre frequency and bandwidth of the generated radar waveforms can be arbitrarily configured over a wide range, in this case 40–50 GHz limited only by the electrical amplifiers. The high carrier frequency and large bandwidth enable us to achieve multi-target ranging with a distance resolution of 1.50 cm, velocity measurement with a resolution of 0.067 m s⁻¹ and inverse synthetic aperture radar (ISAR) imaging with a two-dimensional resolution of 1.50 cm × 1.06 cm.

Results

Figure 1 shows the conceptual illustration and working principle of our photonic mmWave radar chip. We use linear frequency-modulated waveform (LFMW), whose frequency varies linearly with time, as radar waveform, since it features high ranging resolution, constant modulus, Doppler tolerance, as well as a straightforward frequency de-chirp process for the echo waveforms that can greatly alleviate the sampling rate requirements of the radar receiver. To generate the mmWave LFMW signal at the transmitter side, we first implement a frequency multiplying module that performs the frequency doubling of a low-frequency microwave LFMW signal. Specifically, an optical carrier with a frequency of f_c is modulated by a CMOS digital-to-analogue converter (DAC)-produced microwave LFMW signal using a first high-speed TFLN amplitude modulator (EOM1). It features an instantaneous frequency of f₁ + kt that linearly changes from f₁ to f₁ + kT with a bandwidth of B = kT, where T is the waveform period. Biasing the EOM1 at the null transmission point leads to a carrier-suppressed double-sideband (CS-DSB) modulation process that projects the input microwave signal into two LFMW optical sidebands with frequencies of f_c + f₁ + kt and f_c − f₁ − kt, respectively. Afterwards, the modulated optical signal is divided into two paths by a 50%:50% multimode interferometer. Optical signal in the upper path is detected by a high-speed PD (PD1) to generate an mmWave radar waveform whose initial frequency (2f₁) and bandwidth (B₂ = 2kT = 2B) are both doubled from the DAC-input electrical signal. The generated radar waveform is then amplified by a trans-impedance amplifier (TIA) and emitted into free space by an antenna. When the emitted radar waveform encounters targets, the waveform will be reflected with a time delay of τ. The reflected echo waveform is collected by a receiving antenna, amplified by a low-noise amplifier (LNA), and sent to the frequency de-chirp module consisting of a second modulator (EOM2) fabricated on the same TFLN chip. The input port of EOM2 is connected to the lower output path of the abovementioned 50%:50% multimode interferometer, therefore featuring two carrier frequencies of f_c + f₁ + kt and f_c − f₁ − kt, which are subsequently modulated by the amplified echo signals with an instantaneous frequency of 2f₁ + 2kt − 2kτ. By setting the EOM2 at the quadrature transmission point, four new optical sidebands are generated, out of which two sidebands are located in the vicinity of the two carriers, at frequencies of f_c + f₁ + kt − 2kτ and f_c − f₁ − kt + 2kτ. This allows us to achieve frequency de-chirp and obtain the low-frequency target information (2kτ) by beating these two relevant sidebands with the two nearby carriers at a low-speed PD (PD2) and further processing using a low-speed analogue-to-digital converter (ADC). Finally, the ranging, velocity and imaging information of the targets under detection are obtained through subsequent data processing (Methods). The right panel of Fig. 1a shows an envisioned application scenario of our compact photonic mmWave radar in future adaptive cruise control, offering high-resolution distance/velocity detection and imaging capabilities simultaneously, which are key to the enhanced safety, perception and decision-making processes.

Fig. 1: Photonic mmWave radar chip and radar waveform generation.

Benefiting from the excellent modulation performance and scalability of the TFLN platform, our compact photonic mmWave radar enables significantly improved overall performance in terms of integration level, radar resolutions and functionalities compared with previous photonic radar demonstrations (Extended Data Tables 1 and 2). Even better radar resolution down to millimetre levels could be realized by further increasing the frequency and bandwidth of the radar waveforms into upper and more practically important bands (for example, 76–79 GHz for vehicular radars). Leveraging capacitively loaded travelling-wave electrodes³⁵, TFLN EOMs could exceed 80 GHz bandwidth on the same wafer-scale fabrication platform (extrapolated from our previous work³⁹). Advanced frequency multiplying architectures, such as frequency quadrupling or octupling using dual-parallel Mach–Zehnder modulators, could be used to generate high-frequency radar waveforms while maintaining low DAC requirements. Sparse stepped-frequency chirps and coherent fusion technology could be used to overcome bandwidth limitations and achieve improved resolution in a practical environment with crowded spectrum occupancy and electromagnetic interference³⁰. The excellent performance and scalability of the TFLN platform together with recent efforts in heterogeneous and hybrid integration^46,47,48,49 have unlocked the possibility of integrating and co-packaging all the used photonic and electrical devices on chip and board levels, leading to the realization of a compact and low-cost integrated photonic mmWave radar full system. The TFLN photonic mmWave radar could provide compact, low-cost and high-resolution solutions for diverse applications such as sensing⁵⁰, imaging³⁰, smart home⁵¹, environmental monitoring⁵² and the seamless integration of communication and radar systems in the anticipated 6G technology era²³.

Methods

Design and fabrication of the devices

Devices are simulated using Ansys Lumerical Mode and High Frequency Simulation Software (Ansys HFSS). A 4-inch TFLN wafer with a 500 nm x-cut device layer from NANOLN is used to fabricate the designed devices. First, an etching hard mask of SiO₂ is deposited on the LNOI surface through plasma-enhanced chemical vapour deposition (PECVD). An ASML UV Stepper lithography system (NFF, HKUST) with a resolution of 500 nm patterns waveguides, EOMs and MMI on the 4-inch LNOI wafer die by die (1.5 cm × 1.5 cm). Then, the exposed pattern is transferred to the SiO₂ hard mask using a standard fluorine-based dry etching process and to the TFLN layer with 250 nm waveguide height and 250 nm slab height using an optimized Ar⁺ assistant ICP-RIE process. Afterwards, a second stepper lithography patterns the electrode layer after removing the residual SiO₂ mask. After metal evaporation and lift-off process, ground–signal–ground electrodes with a gap of 5.5 µm are obtained, which can ensure good electro-optic modulation efficiency and low metal-induced optical losses. Finally, the chips are cladded in SiO₂ using PECVD, cleaved and facet polished.

Principles of the photonic mmWave radar

The detailed principles of the photonic mmWave radar are presented. To generate the radar waveform, the EOM1 is driven by a fundamental linear frequency-modulated signal. The output optical signal at the output of EOM1 is expressed as

where E₀ and f_c are the amplitude and frequency of the optical carrier, V₁, f₁ + kt, k, T and B (B = kT) are the amplitude, instantaneous frequency, chirp rate, waveform period and bandwidth of the fundamental linear frequency-modulated signal, V_DC1 is the applied DC bias voltage and V_π is the half-wave voltage of the EOM1. By setting the DC bias of EOM1 at the null transmission point, a CS-DSB modulated optical signal is obtained, which can be written as

where J_n is the n-order first-kind Bessel functions and β₁ = πV₁/V_π is the corresponding microwave signal modulation index of the EOM1. Afterwards, the optical signal is divided into two parts by a one-in-two (50%:50%) multimode interferometer. As shown in the upper path of Fig. 1a, half the optical signal is detected by PD1 to achieve photoelectric conversion. The recovered high-frequency electrical signal in PD1 can be given by

Thus, the initial frequency (f₂ = 2f₁) and bandwidth (B₂ = 2kT = 2B) of the generated radar waveform are doubled from the driving electrical signal. Afterwards, the radar waveform is amplified by a TIA and emitted to free space by an antenna. When the emitted radar waveform meets a target, the waveform will be reflected with a time delay of τ. The reflected echo waveform is collected by a received antenna and amplified by an LNA. By applying the amplified echo waveform to EOM2 and biasing EOM2 at the quadrature transmission point, the optical signal at the output of EOM2 is expressed as

where β₂ = πV₂/V_π and V₂ are the corresponding modulation index of the EOM2 and amplitude of the echo waveform.

Afterwards, the PD2 is used to recover the frequency de-chirped electrical signal, by beating the two optical sidebands at frequencies of f_c + f₁ + kt and f_c + f₁ + kt − 2kτ (or f_c − f₁ − kt + 2kτ and f_c − f₁ − kt), which is given by

The frequency (f_de) of the de-chirped electrical signal is 2kτ, which is proportional to the time delay of the echo waveform. Thus, the range (R₁) between the radar antenna and the detected target can be equal to cτ/2, which is written as

The ranging resolution of the radar can be determined by the 3 dB bandwidth of the de-chirped signal (Δf_3dB). In an ideal case, f_de is equal to the minimum distinguishable frequency difference (f_de = Δf_3dB = f_min = 1/T), so the theoretical ranging resolution (ΔR) can be expressed as

According to the radar equation, the distance between target and radar R can be expressed as

where P_r is the power received by the radar, P_t is the peak transmitter power, G_t is the transmitter gain, σ is the radar cross section and A_e is the effective area of the receiving antenna. The angular resolution, defined as the distance between two targets, can be calculated using the following formula as

where θ represents the antenna beamwidth from commercial antenna datasheet, S_A is the angular resolution as a distance between two targets and R is the slant range aim.

In addition, our radar possesses velocity detection capabilities. When a target moves towards or away from the radar line of sight by a distance of Δr, the received signal experiences an additional delay (Δτ), which is introduced by a Doppler shift compared with the echo signal received when the object is stationary. This additional phase change in the de-chirped electrical signal can be expressed as

where f_rc is the centre frequency of the radar waveform, λ is the wavelength of the transmitted radar signal and T is the period of the chirped transmission signal. The phase exhibits a linear response to small distance changes in the object. The phase difference measured across two consecutive chirps can be used to estimate the velocity of the target. Hence, the estimated velocity, obtained from the measured phase difference, is expressed as

On the basis of the properties of the Fourier transform, the peak phase of the Fourier spectrum represents the initial phase of the signal. However, when measuring velocity with multiple objects using M-period chirps transmitted within a ‘frame’ (T_f = 1/MT), the values at the peak that contain phasor components from different targets can be distinguished by performing an FFT on the sequence corresponding to the range-FFT peaks, commonly known as a Doppler-FFT. Here the value of M is 12,500, which is obtained by dividing the total time width of the signal for a 50 ms by the one period of the signal for a 4 μs. If Δφ equals the minimum distinguishable phase difference (Δφ_min = 2π/M), the detected velocity is referred to as the theoretical velocity resolution (Δv), which can be calculated as

The maximum unambiguous velocity can be calculated as

where f_PRF is the repetition frequency of the radar waveform.

Finally, an ISAR system has been demonstrated using our photonic mmWave radar chip. A turntable model is used to analyse the movement of targets for ISAR imaging. The imaging process involves the relative rotation between the radar and the target, which can be observed in the joint range–Doppler domains. The target that is detected undergoes rotation at an angular velocity of ω_t. To process the de-chirped electrical signal, we use a low-speed OSC that performs multi-period sampling for a duration of T_i, equivalent to one coherent processing interval (CPI). The collected data is then rearranged into a two-dimensional matrix (M × N) of delay time (M) and pulse number (N). By leveraging the intrinsic relationship between delay time and distance, it is possible to achieve distance compression through the application of a Fourier transform on M data points within each received echo, thereby generating a range envelope. Consequently, the resonant peak discerned within the range envelope accurately denotes the positional distance of the principal scattering point. The distance can be calculated based on the distance formula (equation (6)). The position of the amplitude peak in the distance envelope represents the distance location of the main scattering points. The translational component of the target with respect to the radar is useless for ISAR imaging, so motion compensation (including distance alignment and phase compensation) on the range compressed data to eliminate phase terms is necessarily required. At each distance unit, we conduct a Fourier transform on the reflected signals of N pulses, generating an N-point Doppler frequency domain and achieving azimuthal compression. As a result, through the initial FFT process of the 2D-FFT, the complex information obtained is equivalent to that acquired through physical I/Q mixing⁵³. By applying Fourier transforms to both the distance and azimuthal dimensions, we can obtain the image of the measured target. The resolution in the distance dimension remains the same as that in ranging detection, while the resolution in the azimuthal dimension (ΔA) is expressed as

where f_a is equal to the minimum distinguishable frequency difference (f_min = 1/T_i), f_rc is the centre frequency of the radar waveform and Δθ is the accumulated angle during radar detection. The azimuthal resolution is determined by the total number of pulses (N) within the CPI.

Principle of the TFLN-based frequency doubling module with limited Y-branch optical splitting ratio

Owing to fabrication errors, the splitting ratio of the Y-branch in the intensity modulator deviates from the ideal value of 50%:50%. Here we set g and 1-g as the optical splitting ratio of the Y-branch. By biasing the intensity modulator at the null transmission point and considering small-signal modulation, the output optical signal can be expressed as

where E₀ and f_c are the amplitude and frequency of the optical carrier, V₁, f₁ + kt and k are the amplitude, instantaneous frequency and chirp rate of the applied LFMW, V_π is the half-wave voltage of the modulator, V_DC1 = V_π is the applied DC bias voltage, J_n is the nth order Bessel functions of the first kind and β₁ = πV₁/V_π is the corresponding RF modulation index. As shown in equation (15), the output consists of two first-order sidebands and the residual optical carrier. Subsequently, the optical signal is detected by a PD to achieve photoelectric conversion. The generated radar waveform is given by

where the first and second terms represent the direct current component, and the third term is the generated frequency doubled LFMW radar signal. Thus, the residual optical carrier does not generate harmonics. We have conducted additional simulations using MATLAB to evaluate the effect of finite sideband-to-carrier suppression ratios in practical devices, for example, owing to unideal Y-branch splitting ratios, as illustrated in Extended Data Fig. 2. Even when the sideband-to-carrier suppression ratio decreases from 24 dB to 9 dB, the fundamental frequency tone remains suppressed in the output electrical spectra. The higher-order harmonics (80–100 GHz) are automatically filtered out owing to the 50 GHz bandwidth of the photodetector.

Characterization of the devices on the TFLN platform

To test the optical performance of the fabricated TFLN chips, optical input signal from a tunable telecom laser source (Santec TSL-550) is coupled to the chip using a lensed fibre. The output optical signal is collected by another lensed fibre and sent to a low-speed PD (125 MHz New Focus 1811). The optical loss of the TFLN waveguides is estimated by measuring the optical transmission spectrum of a racetrack resonator and fitting with a Lorentzian function. The fabrication process of this chip is the same as our previous work and the waveguide propagation loss is <0.1 dB cm⁻¹ (ref. ³⁹).

To measure the half-wave voltage (V_π) of EOMs, a kilohertz electrical triangular waveform generated from an arbitrary-waveform generator (AWG, RIGOL DG4102) is applied to the ground–signal–ground electrodes of the EOM through a probe (GGB Industries, 50 GHz). The output optical signal of the EOM is detected using the same low-speed PD and monitored using a low-speed oscilloscope (RIGOL DS6104). For electro-optic S₂₁ response measurements, a frequency sweeping electrical signal generated from a 53 GHz vector network analyser (VNA, Keysight E5080B) is sent to the EOM with a 50 Ω load. A high-speed PD (XPDV2120R) is used to detect the modulated optical signal. The recovered electrical signal is sent back to the input port of the VNA. After calibrating the frequency responses of the probe, electrical cables and PD, the S₂₁ frequency response of the EOM can be obtained, showing 3 dB bandwidths larger than 50 GHz in this case. Compared with discrete photonic radars, we not only reduce the size of the modulation block from 2 × 135.0 mm × 11.4 mm to 15 mm × 1.5 mm but more importantly substantially improve the modulation half-wave voltage and bandwidth performances, as detailed in Fig. 1. Our modulator features significantly lowered RF V_π and larger EO bandwidths compared with commercial bulk lithium niobate modulators (Thorlabs, LNA6213, LNA 6112 and EOSPACE AZ-DV5-60), as shown in Extended Data Fig. 1. These metrics directly translate into the power consumption, frequency band, resolution and size of the final photonic radars.

Ranging, velocity and imaging measurement of the photonic mmWave radar

In our proof-of-concept radar detection experiments, a continuous wave optical carrier emitted from the tunable laser is first amplified by an erbium-doped fibre amplifier (HaoMinOE EDFA-C-4) and coupled to the TFLN radar chip through a lensed fibre. Benefiting from the high-power handling capability of TFLN, the high input optical power can improve the SNR of the modulated output optical signal from the chip. A fibre polarization controller (FPC) is used to tune the input optical signal to TE mode for the largest EO modulation efficiency. The fundamental LFMW signal (with instantaneous frequency linearly increasing from 20 GHz to 25 GHz within 4 μs) is generated from a high-speed AWG (Keysight M8196A, 33 GHz), amplified by an electrical power amplifier (Fairview Microwave PE15A4021), combined with a DC voltage through a bias-tee (Marki Microwave BT-0050), and used to drive EOM1 via the high-speed probe. The DC voltage is used to bias EOM1 at the null transmission point for the CS-DSB modulation scheme here. The optical spectrum of the CS-DSB modulated signal is monitored using an OSA (YOKOGAWA AQ6370D). At the output side of the photonic mmWave radar chip, a lensed fibre array is used to collect the output optical signals from EOM1 and EOM2, respectively. The upper output path of EOM1 is amplified and detected by the high-speed PD1 to generate mmWave radar waveform, which can be monitored using an electrical spectrum analyser (Agilent N9030A) and analysed using a high-speed oscilloscope (Keysight UXR0404AP). In actual radar testing, the generated radar signal is amplified by two-stage electrical amplifiers (Centellax OA4MVM3) before being emitted to free space through a horn antenna (SAGE Millimeter WR-22 SAZ-2410-22-S1). The reflected echo waveform is collected by another horn antenna of the same type, amplified first by an LNA (SHF S807) and then a power amplifier (Centellax OA4MVM3), before being used to drive EOM2 through another high-speed probe. The optical output of EOM2 is amplified and detected by the low-speed PD2. An oscilloscope (DSO-X 91604A) is used to capture the recovered electrical signal with a sampling rate of 1 GSa s⁻¹. The low-speed PD2 and oscilloscope in the receiver possesses natural filtering characteristics as it cannot respond to electrical signals beyond its bandwidth. Consequently, undesired harmonics are naturally eliminated. In addition, a low-pass electrical filter can be incorporated into the radar receiver to further ensure effective harmonic filtering.

For ranging experiments, metallic plates with a size of 7 cm × 10 cm are used as targets, which are placed at different positions with respect to the antenna. The reflected surface is perpendicular to the ranging direction, which ensures that the reflection surface has a well-defined distance from the antenna, ideally unaffected by the cross-section size. After an FFT process in real time (with a latency of 40 ms in a personal computer), we can obtain the distance between the target and radar. The average range resolution (3 dB bandwidth) of the ranging measurement is 1.71 cm (285 kHz), which is slightly larger than the theoretical value of 1.50 cm (250 kHz). This discrepancy may be due to the imperfect perpendicularity of the metallic plate, causing a slight superposition of multiple adjacent target points within the main lobe of the de-chirped signal spectrum. By integrating artificial intelligence technologies such as convolutional neural networks with radar processing algorithms, it is possible to enhance both the refresh rate and detection accuracy even further. In addition, upgrading the computer platform configuration can also lead to further improvements in processing speed. The detection distance of a radar system is directly proportional to the power of its emission, radar cross-section and radar antenna gain (equation (8)). On the basis of the test results, we can derive a fitting curve related to distance and received power, as shown in Extended Data Fig. 3a. Extended Data Fig. 3b shows the SNR of our measurement results during different distances between the target and antenna. By extrapolating the fitted line to the point, we estimate the maximum detection distance of our radar to be 17.1 m at the SNR approaching zero and 9.6 m at an SNR of 10 dB under our current equipment setting, mainly limited by the electrical amplifiers, antennas as well as environmental conditions. We could improve these points and achieve larger ranges of hundreds of metres by increasing the intensity of the radar transmitted and reflected signals. Moreover, the angular resolution can be calculated from equation (9), which implies that the angular resolutions are 71 cm and 81 cm on the E-plane and H-plane at a range of 420 cm, respectively.

For velocity measurements, a toy balanced car with a size of 9 cm × 6 cm is used as the target, whose velocity can be programmed by a field programmable gate array. By sampling 50 ms of the recovered electrical signal using an oscilloscope, velocity information is obtained after Doppler-FFT, corresponding to a velocity resolution of 6.7 cm s⁻¹. For ISAR demonstration, the targets are placed on a home-made turntable with a set rotate speed of 1 round per second. The images of the detected targets are obtained with a two-dimensional resolution of 1.50 cm × 1.06 cm after a two-dimensional FFT process to the collected 50 ms recovered electrical signal. To show the azimuthal resolution clearly, we illustrated the measured ISAR results of a single metal corner reflector in Extended Data Fig. 4, which includes the ISAR two-dimensional imaging, the power distribution along an azimuthal slice and the corresponding zoomed-in views, respectively. From Extended Data Fig. 4d, it could be seen that the measured azimuthal resolution at 3 dB bandwidth is 1.06 cm.