Electro-optic (EO) modulators have low half-wave voltage, large bandwidth, and good signal quality. They are widely used as core devices in microwave photonics (MWP). Among them, the thin-film lithium niobate (TFLN) photonic platform has attracted widespread attention from scholars due to its obvious advantages. In recent years, people have conducted research on the preparation of thin film lithium niobate EO modulators. However, most modulator chips are not packaged but are directly coupled for testing through high-frequency probes. There are few reports on MWP systems built with TFLN-based modulator chips.

The Doppler frequency shift (DFS) is a classic MWP system^[1–4]. Tang et al. attempted to integrate two dual-drive Mach–Zehnder modulators (DDMZMs) and micro-ring resonators (MRRs) on a silicon chip to achieve simultaneous measurement of DFS and angle of arrival (AOA)^[5]. Zhang et al. also used a silicon-based dual-parallel Mach–Zehnder modulator (DPMZM) and external wavelength division multiplexer (WDM) system to achieve simultaneous measurement of DFS and AOA^[6]. However, due to the lack of the Pockels effect, pure silicon-based modulators rely on carrier injection or loss in the p-n junction, resulting in an inherent trade-off between modulation efficiency and optical loss^[7]. The invention of an ultra-low-loss TFLN waveguide platform can overcome this fundamental limitation^[8–11]. EO modulators based on etched ridge waveguides on the TFLN platform have the advantages of small size, low radio frequency (RF) drive voltage, and high modulation bandwidth^[12–14]. These advantages of the DFS measurement system based on the TFLN platform provide ease of integration, low power consumption, and wide operating frequency bandwidth.

In this paper, we develop a metalized packaged TFLN DPMZM based on lithium niobate thin-film materials. The device we manufactured achieves a 3 dB electro–electro (EE) bandwidth of 29 GHz and Vπ of 6 V. Based on the manufactured TFLN DPMZM, we build a DFS measurement system. In the proposed measurement system, light generated by external light sources is coupled into the chip. The echo signal received by the radar is fed to one of the RF inputs of the DPMZM, and a reference signal with a frequency of 2 MHz higher than the transmitted signal is fed to the other RF port of the DPMZM. After modulation by two RF signals, the light wave is coupled out of the chip and sent to the low-speed photodiode. By measuring the frequency of the output beat signal, DFS without direction ambiguity can be obtained.

As shown in Figs. 1 and 2, the DPMZM consists of two push–pull sub-MZMs connected in parallel. The device contains a forward tapered spatial-mode filter section, 1×2 multimode interference (MMI) couplers, a 6-mm-long single-drive push–pull modulation section, and heating electrodes. The forward tapered waveguide is designed to couple the light in/out of the chip and filter higher-order modes. 3 dB MMI is intended for beam splitting and combining. The traveling-wave electrodes (TWEs) in the modulation area are arranged in a tight G-S-G-G-S-G layout, with the input/output section of the TWEs straddling the ridge optical waveguide and extending to the edge of the chip for subsequent packaging. To minimize the excitation of higher-order modes, the MMI incoming and outgoing optical waveguides adopted an adiabatically tapered mode size converter, and the S-bend waveguide profile is either a slow cosine or a larger radius arc. Meanwhile, in an effort to reduce the imbalance between the two arms of the waveguide and to ensure that the chip is regulating the direct current (DC) bias voltage, we have made extra backups for the hot electrodes, despite only three DC bias controls being required for practical use.

A cross-sectional view of the DPMZM modulation zone is shown in Fig. 3(a). The device is fabricated on a commercial x-cut lithium niobate on insulator (LNOI) wafer with a 525-µm-thick Si substrate, a 2-µm-thick buried oxide (BOX), and a 600-nm-thick TFLN layer. The waveguide has a ridge height of 300 nm (hr=300nm), half of the TFLN total thickness. The light propagation direction in the modulation section is along the y-axis of the lithium niobate (LN) to utilize the maximum electro-optical coefficient γ33. The effective indices versus waveguide widths for different modes is shown in Fig. 3(c). In conjunction with the actual etching, the simulated waveguide sidewall inclination is set to 75°. Considering the scattering loss caused by the rough sidewalls due to etching, the waveguide should be slightly wider to reduce the overlap of the sidewalls with the modes. The top width of the waveguide is finally set to w=1µm, and its optical mode field is shown in Fig. 3(b).

Compared to directional couplers and Y-branches, the 3 dB MMI coupler for beam splitting/combining has larger manufacturing tolerances, wider operating bandwidths, and insensitive polarization. As shown in Fig. 4, the key parameters, length and width, of the MMI are simulated and designed. The width of the multimode waveguide is set as 8 µm to balance the loss of high-order modes and footprint. Lumerical simulation shows when the multimode region is excited to produce the first-order mode, the distance of mode separation is 4.22 µm, and the length required for self-image is 49 µm.

The design and optimization of the signal electrode require a combination of 50 Ω impedance matching, optical loss, microwave transmission loss, and modulation efficiency. We set signal electrode width ws=14µm, electrode space wg=7µm, and electrode height hAu=0.9µm. Figure 3(d) indicates the electric field distribution for an applied DC voltage of 1 V. Most of the electric field is confined between the electrodes.

All electrodes are eventually bent at 90° and gradually widened toward the side of the chip to facilitate subsequent packaging. To allow the waveguide to pass under the electrodes and minimize the metal absorption loss of the optical signal, an 800 nm SiO2 cladding is introduced between the waveguide and the electrodes. The short TWEs result in lower RF losses but higher half-wave voltage (Vπ). In this work, the TWEs are designed to be 6 mm to balance the bandwidth with Vπ.

The simplified process flow of the proposed device is shown in Fig. 5. Cr is used as a mask to pattern the optical waveguide via electron-beam lithography and dry-etching. Then an 800-nm-thick SiO2 buffer layer is deposited on top of the TFLN using plasma-enhanced chemical vapor deposition (PECVD), followed by the electron-beam evaporation of NiCr and Au onto the sample, and finally the NiCr and Au are lift-off to form the electrodes. The final device footprint is only 15mm×2mm, significantly smaller than commercial devices (EOSPACE AZ-DV5-65, 105mm×17mm).

Figure 6 presents a conceptual diagram of the DFS measurement system based on an integrated LNOI. An external optical carrier from the LD is launched into the on-chip DPMZM after the PC has polarized it. The DPMZM contains two push–pull sub-MZMs (MZM1 and MZM2) whose bias states are controlled by the DC voltages VDC1 and VDC2, respectively. VDC3 is employed to regulate the phase difference between the two sub-MZMs. The echo signal and the reference signal are injected directly into the RF ports of MZM1 and MZM2, respectively. Both sub-MZMs are biased at the minimum point and form a carrier suppression double-sideband (CS-DSB modulation). The modulated optical signal is output from the chip and amplified by an EDFA before entering a low-speed PD. The peak power of the output signal can be observed by the ESA.

The continuous optical carrier output from the LD can be expressed as E(t)=E0exp(jωct),(1)where E0 and ωc are the amplification and angular frequency of the optical wave, respectively.

The reference signal emitted by the microwave source and the echo signal received by the antenna can be written as VRF(t)=Vrsin(ωrt),(2)VEcho(t)=Vesin(ωet),(3)where Vr and Ve are the amplitude voltages of the reference signal and echo signal, respectively.

As mentioned before, sub-MZM1 and sub-MZM2 are fed with the echo signal and the reference signal, respectively, so that the output of the two sub-MZMs and the optical field can be described respectively as E1(t)=12E0ejωct[ejβesin(ωet)+e−jβesin(ωet)ejφ1],(4)E2(t)=12E0ejωct[ejβrsin(ωrt)+e−jβrsin(ωrt)ejφ2]ejφ3,(5)where βe, βr are the RF modulation indices of DPMZM, which are equal to πVe/Vπ and πVr/Vπ, respectively. Also, φ1=πVDC1/Vπ and φ2=πVDC2/Vπ are the phase shifts caused by the DC bias of MZM1 and MZM2, respectively, and Vπ is the half-wave voltage of the DPMZM. Setting φ1=φ2=π, using the Jacobi–Anger expansion and ignoring higher-order sidebands, Eqs. (6) and (7) can be rewritten as E1(t)=12E0ejωct×J1(βe)[ejωet+e−jωet],(6)E2(t)=12E0ejωct×J1(βr)[ejωrt+e−jωrt]ejφ3,(7)where Jn(·) denotes the nth order of the first kind Bessel function. Also, φ3=πVDC3/Vπ is the optical phase shift introduced between the two sub-MZMs in the DPMZM. Setting φ3=0, the optical signal output by the DPMZM can be derived as $$\begin{gathered} Eout(t)=12E0ejωct×[J1(βe)(ejωet+e−jωet) \\ +J1(βr)(ejωrt+e−jωrt)]. \end{gathered}$$(8)

The system adopts a low-speed detector, so the higher-order terms ωe and ωr can be directly ignored. Filtering the DC term, the detected signal can be expressed as I(t)=Eout(t)·Eout(t)*,(9)I(t)∝14E02η×4J1(βe)J1(βr)cos[(ωr−ωe)t],(10)where η is the PD responsivity. The DFS value and direction can be obtained by measuring the frequency of the output electrical signal. In most cases, the DFS value is within ±1MHz^[15,16], so the reference frequency is set to the transmit frequency plus 2 MHz to resolve the DFS directional ambiguity in the measurement. The relationship between the two is given by fDFS=2MHz−(fr−fe).(11)

Figure 7 illustrates a photograph of the packaged integrated LNOI DPMZM. The fiber-to-chip coupling is applied to the light package. For the electrical package, gold wire bonds the pads of the electrodes to the bias circuit board or the RF circuit board, which in turn is soldered to the DC pins or RF header pins.

We use the vector network analyzer (VNA, ROHDE&SCHWARZ, ZNB 40) to describe the electrical response of the packaged TFLN DPMZM. The RF cables are first normalized by means of SOLT calibration. The detector used in the tests is a FINISAR XPDV3120R Ultrafast Photodetector with a bandwidth of 70 GHz. The S21 parameters in Fig. 8(a) is normalized with a reference point of 100 kHz, and the test displays a 3 dB EE bandwidth of ∼29GHz for both sub-MZMs, indicating that the two arms of the prepared DPMZM have good agreement. There is a noticeable undulation in the low-frequency section, which is due to high reflection from an impedance mismatch at low-frequency. Notably, the modulator bandwidth is 17 GHz higher than that of a comparable Si-based integrated measurement system^[5]. In addition, the S11 parameters are below −10dB at frequencies up to 40 GHz, suggesting good impedance matching, which is important for the reduction of RF power loss caused by reflections among the testing system, the chip, and the 50 Ω resistor. The EE bandwidth of the device can be further increased if the TWEs can be further optimized with a T-shaped segmented electrode design to achieve microwave-optical matching.

The frequency-dependent RF Vπ is a major consideration in power efficiency. We measured the low-frequency Vπ of two sub-MZMs with a direct scan of a triangular wave at 100 kHz and 10 Vpp. The sub-MZMs are in a single drive push–pull configuration so that the applied voltage caused a positive phase shift in one arm and a negative phase shift in the other arm. The Vπ of both MZMs is approximately 6 V. The half-wave voltage is not ideal enough due to the introduction of the isolation layer to simplify the manufacturing process, but it is sufficient for scenarios where the half-wave voltage is not in high demand. To further reduce the half-wave voltage, the electrodes can be placed in the same plane as the waveguide (without buffer layer), which improves the overlap integral between the optical and electric fields. However, in this case, it is difficult for the electrode to span the waveguide structure, and thus slow-varying slope SiO2 cladding structures need to be prepared. Figure 8(b) illustrates the Vπ of one of the sub-MZMs.

An experimental setup based on the one shown in Fig. 6 is built to verify the proposed scheme. The laser source (Santec TSL-550) emits linearly polarized light as an optical carrier at 1550 nm with a power of 10 dBm. The optical carrier is polarized by a PC and injected into the packaged DPMZM. External microwave source 1 (MSG-1, KEYSIGHT E8267D) generates a microwave signal with a frequency of 15GHz+2MHz and a power of 10 dBm as the reference signal; external microwave source 2 (MSG-2, ROHDE & SCHWARZ SMA100B) generates a microwave signal with a frequency of 15GHz±1MHz and a power of 10 dBm as the echo signal. In the practical application, the antenna receives a weak echo signal. At the same time, the TFLN modulator also features a large fiber-to-chip coupling loss due to the mode-field mismatch. To improve the sensitivity of the whole measurement system, a multistage low-noise RF amplifier is first needed to amplify the echo signal^[17]. Here, the power of the echo signal is set to 10 dBm to simulate this situation. The reference signal and the echo signal are fed into sub-MZM1 and sub-MZM2, respectively. To determine the bias voltage, the output spectra of the DPMZM when the RF signal is separately fed into the two sub-MZMs are observed by a spectrum analyzer (OSA, YOKOGAWA, AQ6370D) with a resolution of 0.02 nm. The thermal electrodes are adjusted to maximize the sideband power and suppress the carrier as much as possible. As shown in Fig. 9, even though the carrier is in the optimally suppressed state, the carrier is still about 17 dBm above the sideband. This is because there is no RF signal fed into the other sub-MZM. In the process of determining the optimal bias points of the two sub-MZMs (acquiring the two spectral traces as shown in Fig. 9), the thermal bias points of the main MZMs are kept the same, which is consistent with practical use. In practical use, precise control of the two sub-MZMs to operate at the ideal CS-DSB point is not required. The signal output from the DPMZM is amplified by an EDFA and sent to a low-speed PD (THORLABS, DET01CFC/M) with a bandwidth of 2 GHz. The DFS is obtained by analyzing the frequency value of the peak signal from the ESA.

Figure 10 depicts the electrical spectrum of the down-converted signal output from the ESA when both the echo and the reference signals are set to 10 dBm. The frequency of the detection signal in Fig. 10(a) is 2.2 MHz, which implies a DFS of −0.2MHz. This corresponds to a negative direction of the DFS, where the target is moving away from the receiver. In contrast, the frequency of the detected signal in Fig. 10(b) is 1.8 MHz, which means that the target is approaching the receiver with a DFS of +0.2MHz.

To minimize the inherent error, the MSG set at a specific frequency is first connected directly to the ESA, and the difference between the measured value of the spectrometer and the set value of the microwave source is taken as the inherent error and subtracted in subsequent measurements. The frequency of the echo signal then varies from 15 GHz − 100 kHz to 15 GHz + 100 kHz with a step of 10 kHz. During this experiment, the ESA operating with 1 Hz resolution bandwidth (RBW), 1 Hz video bandwidth (VBW), and 100 Hz span is utilized to monitor the spectrum. Figure 11(a) displays the measurement error of the DFS within ±0.3Hz, and the measured values agree well with the theoretical values. The actual measured values are plotted with green lines and squares, and the measurement errors are plotted with blue lines and dots, respectively. Here, the measured DFSs are the specified frequency differences among MSG-1, MSG-2, and 2 MHz.

Next, the transmit signal is tuned from 5 to 30 GHz with a 5 GHz step, and the reference signal frequency is set to the corresponding transmit signal frequency plus of 2 MHz, which is used to characterize the DFS measurement error at different transmit signal frequencies. The reference signal can be generated by mixing the transmit signal and a 2 MHz low-frequency signal through an electrical mixer. Figure 11(b) demonstrates the measurement error as a function of the transmit signal frequency when the DFS is 1, 0, and −1MHz. As the frequency of the transmit signal varies, the range of the measurement error remains within ±0.6Hz. The DFS error is mainly influenced by the accuracy of the equipment, so the measured DFS error is independent on the carrier frequency. Frequency measurement errors can be further minimized using a higher-resolution ESA. In a practical application, a better choice is to utilize digital signal processing (DSP) methods to obtain DFS signals whose inherent error is related to the sampling rate of the analog-to-digital converter (ADC).

Adequate experimental results show that our system has good stability and robustness across different transmit signal frequency bands, and the system based on the TFLN DPMZM has a large operating bandwidth. As shown in Table 1, comparing the reported work with the proposed method in this paper, the proposed method realizes DFS measurement based on TFLN for the first time and has the advantages of simple structure and large measurement range. For example, Refs. [15,16] give a simple fiber optic link including an LD, an optical modulator, a filter, and two PDs. This scheme uses only the frequency of the low-frequency electrical signal at the output of one of the PDs to compute the DFS, which is simple in structure, has a large operating bandwidth, and has good robustness. However, the modulator used in this scheme contains a 90° polarization rotator (PR) structure, and despite some related work^[11,12], on-chip control of the polarization state of light is difficult to achieve. Reference [5] demonstrates a simultaneous measurement system for AOA and DFS by connecting two silicon dual-drive MZMs in series and inserting a MRR between them. To stabilize the resonant frequency, the introduced micro-ring structure requires precise control of the coupling coefficient between the ring and the straight waveguide, which can lead to a more difficult control of the system. Based on a silicon DPMZM containing a thermo-optical power splitter (TOPS), two OBPFs, and two PDs, Ref. [6] realizes simultaneous measurements of DFS and AOA. To ensure accurate power-phase mapping, the scheme requires careful control of the homogeneity of the two branches. For DFS measurements, the direction is determined by comparing the phase relationship between the two output IF signals, and the comparison of this phase relationship needs to specify the bias phase of the parent-MZM, which is difficult in practical tests. In contrast, our proposed DFS measurement scheme based on DPMZM on the TFLN platform can simultaneously measure the value and direction of DFS using only a DPMZM and a low-speed PD. Our scheme does not require complex polarization multiplexing structures or additional resonant devices or filter pieces, which reduces system complexity and improves stability, making the measurement more intuitive and convenient, and more suitable for further on-chip integration.

In summary, we have prepared a DPMZM based on the TFLN platform with a complete electrical and optical package. The presented device features a large 3 dB bandwidth of ∼29GHz and a half-wave voltage of ∼6V. Afterward, a MWP system that can measure both DFS value and direction is constructed based on the integrated LNOI DPMZM. In the validation experiments, the problem of ambiguous DFS direction measurements is solved by introducing a reference signal. The DFS value can be measured by the beat frequency of the reference signal and the echo signal. The measurement error is within ±0.3Hz in the range of ±100kHz at a carrier frequency of 15 GHz. The DFS measurement error is also less than ±0.6Hz with the carrier frequency ranging from 5 to 30 GHz. To the best of our knowledge, this is the first DFS measurement system based on an integrated LN circuit. Our scheme provides a feasible way to achieve high-performance monolithic integration of DFS measurement systems. The success of this experiment marks a key step in the practical application of TFLN modulators in MWP.