Widely tunable Ka-band optoelectronic oscillator integrated on thin film lithium niobate platform

The generation of high frequency microwave signals with wide tenability and low phase noise is crucial for applications including communications, radar systems, and electronic warfare.^1–3 Optoelectronic oscillators (OEOs) have shown great promise for producing such frequency-tunable signals.^4–11 Several frequency-tunable photonic integrated OEOs based on different material platforms have been realized, which are superior in size, weight, power consumption, and cost. For example, an electrical bandpass filter (EBF)-based frequency-tunable OEO¹² in an indium phosphide (InP) platform integrating a directly modulated laser (DML), an optical delay line (ODL), and a photodetector (PD) has been demonstrated, but unfortunately with a tuning range extremely limited due to the poor performance of the EBF. Microwave photonic filters (MPFs) have wideband tenability and have therefore been used to realize a widely tunable OEO¹³ on a silicon on insulator (SOI) platform by integrating a phase modulator (PM), a micro-disk resonator (MDR), and a PD. In addition, tunable parity-time (PT)-symmetric OEOs by integrating a micro-ring resonator (MRR) and a tunable Mach–Zehnder interferometer (MZI)¹⁴ or by using an integrated mode-locked laser (MLL)¹⁵ fabricated on an InP platform have also been reported. Finally, a hybrid-integrated tunable OEO¹⁶ including a distributed feedback (DFB) laser and a silicon photonic chip has been realized. Unfortunately, in all the above-mentioned photonic integrated OEOs implemented using various material platforms, oscillation frequencies are all below 20 GHz, likely due to the limitations of the material platforms.¹⁷ 

Here, we experimentally demonstrate an OEO operating at Ka-band and widely tunable from 20 GHz to more than 35 GHz, with the key components photonically integrated on the TFLN platform. In particular, the fabricated photonic integrated chip (PIC) includes an MZI, an MZM, and an add-drop MRR, with the MZM and the MRR placed on one of the MZI arms. The TFLN MZM relies on the strong electro-optic coefficient for efficient modulation and has a modulation bandwidth of >70 GHz and a V_π down to 1.7 V, as previously reported.^18–20 The TFLN MRR is used for selecting a single modulation sideband to beat with the optical carrier from the other arm of the MZI, while eliminating the optical carrier and other modulation sidebands generated by the MZM. Therefore, the oscillation frequency is determined by the spacing between the optical carrier and the selected sideband, with a frequency tuning range determined by the free spectral range (FSR) of the MRR. By tuning the resonances of the MRR, OEO oscillation frequencies from 20 to 35 GHz are realized, with measured phase noises of approximately −85 and −116 dBc/Hz at offset frequencies of 10 and 100 kHz from the OEO oscillation frequency, respectively. In comparison to the photonic integrated OEOs (without using long optical fibers) implemented on other material platforms,^12–14 our PIC-based OEO boosts much higher oscillation frequency, much wider tuning range, but lower phase noise, due to the high modulation bandwidth of the MZM, the wide tuning range, and the high Q of the MRR resulting from the superior material properties of the TFLN. Our research demonstrates not only the feasibility of a unique OEO configuration but also the potential of fabricating fully integrated OEO PICs, especially considering the recent advances in integrating lasers and PDs in TFLN circuits.^21–24 

Figure 1 depicts the schematic of the widely tunable TFLN OEO, realized by a PIC incorporating an MZI, an MZM, and an add-drop MRR, with the MZM and the MRR placed on one of the MZI arms. A CW light provided by a laser (HAN'S RAYPRO) used as an optical carrier is separated into two arms; 90% of the light is coupled to the lower arm containing the MZM and the add-drop MRR, while the remaining 10% goes to the upper arm. A polarization controller (PC) is used to adjust the light entering the PIC to be aligned with the TE mode of the PIC. The MZM is used to modulate the optical carrier to generate modulation sidebands. The erbium-doped optical fiber amplifier (EDFA) (Amonics: AEDFA-PKT-DWDM-15-B-FA) is used for compensating the optical loss so that the optical power into the PD (Finisar: XPDV2320R) remains constant around 9 dBm, which is close to the saturation power of the PD. In addition, the EDFA also functions as a power stabilizer to compensate for the coupling instability due to the mechanical vibrations and drift of our coupling stage because it is operated at saturation.

Initially, the MZM output signal grows from the noise within the OEO loop. The MRR serves as an optical bandpass filter, selecting a sideband of the modulated signal to pass with a minimum loss, which can beat with the un-modulated optical carrier from the upper arm in the PD. The beat signal then feeds back to the MZM before being amplified by the low noise amplifier (LNA) (Talent Microwave) to enhance the selected sideband, which leads to the buildup of the oscillation after the beat signal goes through the loop multiple times if the small signal open loop (SSOL) gain is larger than unity.⁴ The generated RF frequency is determined by the frequency spacing between the optical carrier and one of the resonances of the MRR.²⁵ By thermally tuning the resonances of MRR, the oscillation frequency of the OEO can be tuned accordingly. In contrast, the oscillation frequency of a fixed frequency OEO is determined by the FSR of the add-drop MRR inside the feedback loop, which can hardly be tuned.²⁶ 

Because the transmission loss in the lower arm of the MZI due to the presence of the MZM and MRR is much larger than that in the upper arm, the ratio of the optical power entering the lower arm from the first coupler in the MZI should be much larger than that entering the upper arm to balance out the large loss. Figure 2 shows the simulated results of the SSOL power gain as a function of bias voltage (V_B) with different power splitting ratios, indicating that the SSOL power gain is maximized when the MZM is biased near V_π, with a coupling ratio of 90% to 10%. Such a coupling ratio is what we implemented in the PIC shown in Fig. 1.

The PIC for the ultra-wide range frequency-tunable OEO was designed and fabricated on an X-cut TFLN platform with a device layer of 360 nm, as shown in Fig. 3(a). First, we patterned and defined all optical waveguides through e-beam lithography (EBL) and inductively coupled plasma (ICP) etching processes. Figure 3(b) (left) shows the SEM image of the etched TFLN waveguide and its sidewalls. Subsequently, a 1-μm-thick SiO₂ layer was deposited by plasma enhanced chemical vapor deposition (PECVD). Next, we used e-beam evaporation to fabricate the NiCr load with a thickness of 0.2 μm and the gold layer with a thickness of 0.9 μm. A lift-off process was followed to fabricate the capacitance-loaded traveling-wave electrodes (CL-TWEs). Figure 3(b) (right) shows the SEM image of the zoom-in-view of CL-TWEs. Here, CL-TWEs³⁰ with air bridges were used in TFLN modulators to reduce the geometric length of the device while maintaining a low driving voltage, as shown in Fig. 3(a). The thickness of the silicon oxide above and below the electrode was both 1 μm.

The measured 3 dB EO modulation bandwidth, V_π, and the extinction ratio (ER) of TFLN MZM, with a total modulation length of 2.1 cm, are shown in Figs. 3(d) and 3(e), with values of 38.3 GHz, 1.7 V, and 40 dB, respectively. This large 3 dB EO bandwidth ensures the OEO can generate high frequency signals in the Ka-band. The V_π is significantly lower compared to the typical value of >4 V of the commercial MZMs fabricated with bulk LN,³¹ which will help to relax the required LNA gain for reducing the heat, power consumption, and cost, and even completely eliminate the need for an LNA if the optical power entering the PD is sufficiently high, thus removing the LNA flicker noise contribution to the OEO phase noise.⁴ 

Figure 3(c) shows the SEM image of the zoom-in-view of the coupling region of the MRR. The ring waveguide width (W₁), the bus waveguide width (W₂), and the gap between them (W₃) were 2.2, 1.38, and 0.8 μm, respectively. The measured 3 dB bandwidth, FSR, ER, Q, and thermal tuning efficiency of the add-drop MRR, corresponding to the circumference of 4.5 mm, are shown in Figs. 3(f) and 3(g), with values of 650 MHz, 70.6 GHz, 29 dB, 3 × 10⁵, and 7.4 pm/mW (0.92 GHz/mW), respectively.³² Our Q factor is about four times higher than that of the add-drop MRR (Q = 8.1 × 10⁴), fabricated with the silicon platform used in the OEO,²⁵ which is crucial for generating low phase noise RF signals. Reducing the waveguide scattering loss caused by sidewall roughness can further increase the Q. Meanwhile, a large FSR guarantees OEO’s wide range tenability. The jitter shown in Figs. 3(f) and 3(g) below −40 dB originates from the power meter’s noise because the optical powers at these frequencies are below our power meter’s noise floor.

The 3 dB bandwidth and the tuning range of the add-drop MRR determine the 3 dB bandwidth and tuning range of the SSOL power gain of the OEO, further affecting the phase noise and tenability of oscillation signals. In the measurement, a vector network analyzer (VNA) (Agilent Technologies: N5227A) is placed between the MZM and EC to measure the frequency response of the SSOL power gain obtained from the transmission coefficient S₂₁,²⁶ when a scanning RF signal from the VNA output port is fed into the MZM to linearly modulate the optical carrier with a wavelength of 1550.14 nm and a power of 18 dBm, while the output from the EC is sent to the input port of VNA. By thermally tuning the add-drop MRR, a modulation sideband can be selected by one of the resonances of the MRR to beat with the un-modulated optical carrier from the other arm of the MZI to generate the oscillation signal, with the frequency determined by the frequency spacing between the optical carrier and one of the resonances of the MRR. Figure 4(a) shows the measured frequency response of the SSOL gain, with the center frequency tuned from 20 to 35 GHz by thermally tuning the MRR. It can be seen that the 3 dB bandwidth of the SSOL gain curves at 20–34 GHz is ∼667 MHz, almost the same as that of the MRR, indicating that the optical bandwidth of the MRR is effectively preserved in the microwave frequency domain. Note that at the frequency of 35.3 GHz (FSR/2 of the MRR), corresponding to the special case that both the lower and upper modulation sidebands are selected by the MRR, the gain is much higher than that of the rest gain peaks, as anticipated per the discussion in Sec. II. One may also notice that the gain curve broadens at 35 GHz because of the overlap of the gain curves at 35 and 35.6 GHz.

Figure 4(b) shows the comparison of the frequency response of the OEO’s SSOL gain at 30 GHz between the measured (green curve) and the simulated results (red curve). In simulation, the parameters β₀, β₋₁, β₁, and V_B can be calculated based on the measured normalized optical transmission spectrum of the MZM (mulberry curve), the measured optical transmission spectrum of the add-drop MRR (blue curve), the optical carrier (red dashed curve) with a wavelength of 1550.14 nm, and the desired oscillation frequency of 30 GHz, shown in Fig. 4(c). For obtaining 30 GHz oscillation, one of the resonances of the add-drop MRR is tuned to be at 1550.38 nm through thermal tuning, to select the upper sideband of the modulated signal, while the lower sideband at 1549.9 nm is filtered out, as it does not locate at the MRR’s adjacent resonance at 1549.815 nm. Therefore, the parameters β₀, β₋₁, and β₁ can be calculated to be 0.03, 0.04, and 1, respectively. Ideally, in Eq. (5), the MRR can be tuned to completely filter out the optical carrier in the lower arm such that β₀ = 0, but in reality, the calculated result is 0.03 due to the residual higher-order transverse mode in the waveguide of MRR, as shown by the small bump and discontinuity between the two resonances of the MRR in Fig. 4(c), which prevents the optical carrier in the lower arm from being completely filtered out. In addition, the parameter V_B can be calculated to be 0.99 V_π according to the measured normalized optical transmission spectrum of the MZM and the optical carrier at 1550.14 nm. The parameters at other OEO frequencies can be obtained using the same method.

Figure 4(d) represents the measured power gain of the LNA (blue curve) used and the frequency response of the PD (orange curve), respectively. Based on the frequency response of the PD shown in Fig. 4(d) and the frequency response of the MZM shown in Fig. 3(d), the parameter $hωm$ can be calculated to be 0.76. By substituting the experimentally obtained values of β₀, β₋₁, β₁, V_B, and $hωm$ into Eq. (5), the simulated frequency response of SSOL gain at 30 GHz can be obtained, as shown in Fig. 4(b). It can be seen that the simulated and measured curves of SSOL gain at 30 GHz are practically the same, except for the notable difference between the simulated and measured gain values (3.6 vs 2 dB), mainly caused by our inability to accurately determine the optical losses used in the simulation due to the fluctuations of our chip-to-fiber coupling during measurement.

Figure 5(a) shows the OEO’s oscillation frequencies from 20 to 35 GHz as the MRR is thermally tuned, with the side mode suppression ratio (SMSR) remaining constant in the entire tuning range, an important advantage of the photonically tuned OEO. Compared to the tuning range (3–7.4 GHz) of the OEO implemented with the SOI PIC,¹³ the tuning range of our OEO implemented with the TFLN PIC is much wider (15 vs 4.4 GHz). Even a wider frequency tuning range is feasible if microwave components, such as LNAs with wider bandwidths, are available. Figure 5(b) shows the frequency spectrum of the generated RF signal when the OEO is tuned to 30 GHz, with a SMSR of 47 dB and a mode spacing of 8 MHz determined by the total signal path length of all the optical and RF components used in the OEO loop.

The phase noise curves of generated OEO signals at 20, 25, 30, and 35 GHz are measured with an RF spectrum analyzer (RFSA) (Agilent Technologies: PXA N9030A), as shown in Fig. 5(c). It can be seen that the phase noise curves of the signals generated by the OEO at 20, 25, 30, and 35 GHz are practically the same, which remain high and flat at offset frequencies below 10 kHz from their perspective carrier frequencies and can be attributed to the fluctuations of the laser frequency,²⁶ as well as the random mechanical vibrations of the fiber-chip coupling. Nevertheless, when the frequency offsets from the carrier are beyond 10 kHz, the influences of laser fluctuations and the system mechanical vibration are reduced. In particular, the phase noises at 10 kHz frequency offset are all rapidly dropped to −85 dBc/Hz, which are even lower than the X-band OEO integrated on the SOI platform (−80 dBc/Hz@10 kHz for 3–7.4 GHz).¹³ The oscillation frequency is determined by the frequency spacing between the laser and the resonances of the MRR, and is therefore sensitive to the fluctuations of the laser frequency. In Fig. 5(c), the sharp jump of the OEO’s phase noise around a corner offset frequency f_c of 6.5 kHz is likely attributed to the rapid increase in the laser’s linewidth or the laser frequency fluctuation because the laser’s linewidth is a Sigmoid function of measurement duration,³³ which may have a sharp transition around a measurement duration of 1.5 × 10⁻⁴ s (⁠$=1/fc$⁠. However, confirming this hypothesis requires further experimental validation, which exceeds the scope of this paper. The phase noise of OEO can be further reduced by increasing the Q factor of the MRR and using high-performance optical and electrical components, such as lasers with low frequency jitter and relative intensity noise (RIN), and LNA with low phase noise.²⁶ In addition, the proper packaging to minimize the power fluctuation due to vibration of the chip-fiber coupling, the elimination of EDFA, and the inclusion of a feedback loop³⁴ to lock the laser frequency to the resonances of MRR, can also help to reduce the phase noise close to the carrier and improve the frequency stability. Moreover, for achieving long term stability for applications in a communication or radar system, phase-locking³⁵ the OEO to a system reference RF source is generally required. Figure 5(d) shows the optical spectrum measured with an optical spectrum analyzer (Anritsu MS9740A) when the OEO is tuned to 30 GHz, which shows that indeed one sideband of modulated signal is selected by one resonance of MRR to beat with the optical carrier.

It is interesting to take a close look at the special oscillation at 35.3 GHz, corresponding to half of the FSR of the MRR. Figure 6(a) shows that the phase noise of the 35.3 GHz oscillation is −100 dBc/Hz@10 kHz, 15 dB better than those of the oscillations at 20–35 GHz shown in Fig. 5(c), which may be due to the fact the SSOL gain at this oscillation frequency is much higher than those of the rest oscillation frequencies.²⁵ The measured optical spectrum of the OEO at 35.3 GHz is shown in Fig. 6(b), indicating that indeed two modulation sidebands are present. Figure 6(d) shows the frequency spectrum of the generated RF signal when the OEO is tuned to 35.3 GHz, with a SMSR of 48 dB, consistent with that of 47 dB in Fig. 5(b).

Figure 6(c) shows the simulated (red curve) and measured (blue curve) frequency responses of SSOL gain at 35.3 GHz. It can be seen that the simulated and measured SSOL power gain curves are practically the same, validating our theoretical model, despite a slight difference in gain values at 35.3 GHz between them (8.8 vs 7.3 dB) caused by the inaccuracy of the optical power measurement due to the instable chip-to-fiber coupling. In addition, as can be seen in Fig. 4(a), the gain at 35.3 GHz is higher than that at 20–35 GHz, which is due to the fact that at 35.3 GHz, the oscillating signal is the beat of both the upper and lower modulation sidebands with the optical carrier, whereas at 20–35 GHz, the oscillating signal is the beat of a single modulation sideband with the optical carrier. As discussed in Eq. (5), theoretically, the power gain at 35.3 GHz should be 9.6 dB, 6 dB higher than that (3.6 dB) at 20–35 GHz, which is sufficiently close to the simulated gain of 8.8 dB. The difference may be attributed to the lower response of the PD and the MZM at larger RF frequency of 35.3 GHz. In the simulation of Eq. (5), at the oscillation frequency of 35.3 GHz, the parameters β₀, β₋₁, β₁, V_B, and $hωm$ are calculated to be 0.02, 0.99, 1, 0.99 V_π, and 0.67, respectively.

In this paper, a novel Ka-band widely tunable OEO for achieving compact footprint, low cost, and low-power consumption is proposed and experimentally demonstrated, realized with a TFLN PIC by integrating an MZM, a thermally-tunable add-drop MRR, and an MZI, with the MZM and the MRR sequentially situated in one of the MZI arms. The theoretical analysis of the OEO’s SSOL gain is first provided and subsequently verified by experiments. The low V_π of 1.7 V and large 3-dB EO bandwidth of 38.3 GHz of the MZM, fabricated using folded CL-TWE with a total modulation length of 2.1 cm, help to reduce the required LNA gain and enable high frequency oscillation. In addition, the high Q factor of 3 × 10⁵ and the large FSR of 70.6 GHz of the add-drop MRR contribute to realizing oscillation signals with low phase noise and a wide tunable range. Wide range tunable RF signals from 20 to 35 GHz are generated by thermally tuning the add-drop MRR, with measured phase noises of −85 dBc/Hz@10 kHz and −116 dBc/Hz@100 kHz, which are even lower than those of RF signals generated by the frequency tunable OEO of much lower frequency and narrower tuning range (3–7.4 GHz), realized with the PIC on the SOI platform (−80 dBc/Hz@10&100 kHz).¹³ Even a wider frequency tuning range is feasible if microwave components, such as LNAs with wider bandwidths, are available. The phase noise can be reduced by increasing the Q factor of the MRR, and using high-performance optical and electrical components, such as lasers with low frequency jitter and RIN, and LNA with low phase noise.²⁶ In addition, the proper packaging to minimize the power fluctuation due to vibration of the chip-fiber coupling, the elimination of EDFA, and the inclusion of a feedback loop³⁴ to lock the laser frequency to the resonances of MRR, can also help to reduce the phase noise close to the carrier and improve the frequency stability. Our work demonstrates the potential of utilizing the TFLN platform to achieve photonic integrated OEO with high frequency, wide range tenability, and low phase noise. Due to the incompatibility between the TFLN and the metal-oxide-semiconductor (CMOS) technologies, it will be challenging to integrate the RF LNA and other electrical components on the TFLN PIC, but a hybrid-integrated OEO can be achieved by connecting a TFLN PIC with an electronic chip containing an RF amplifier and other components through wire-bonding to microstrip lines.

The present Ka-band OEO has an input optical power to output RF power conversion efficiency on the order of 8.4% (assuming 1% of the RF power is coupled out of the OEO loop), with a total electrical power consumption of 3.34 W (including the power consumptions of all optical and electronic components used in the OEO), while a typical X-band OEO made with all discrete components in our lab has a typical power conversion efficient of 12.8%, consuming a total electric power of 7.32 W (a typical V_π of 4 V for a bulk X-band LN MZM). The power consumption of a Ka-band OEO with discrete components is expected to be much higher considering that the typical V_π of a Ka-band bulk LN MZM is about 7 V. Nevertheless, compared to an X-band OEO, our Ka-band OEO has lower power conversion efficiency mainly due to the large on-chip optical loss of 15 dB and excessive RF cable/connector loss of over 7 dB at 30 GHz, but lower power consumption thanks to the smaller V_π of the integrated TFLN MZM of 1.7 V. By reducing the on-chip optical loss and the RF cable/connector loss, our integrated TFLN OEO can outperform the OEO made with discrete components in both power conversion efficiency and power consumption, which may prove beneficial for applications requiring lower power consumption and high-power conversion efficiency, such as in 5 G communication and radar systems.

Despite the successful demonstration, the challenges of applying the TFLN OEO in real-world applications, such as the 5G communication and radar systems, still remain, which mainly include (1) achieving low phase noise, which may increase system cost and complexity; (2) maintaining long term stability in various environmental conditions and operating reliably over long periods without failure; (3) minimizing power consumption while meeting performance requirements; and (4) overcoming compatibility issues with other systems and standards to increase the degree of integration and standardization of the OEO. These challenges require further technological innovation and engineering optimization to better meet the requirements of 5G communication and radar applications.

This work was supported by the National Key R&D program of China (Grant No. 2019YFA0705000), the Natural Science Foundation of China (Grant No. 62293523), the Advanced Talents Program of Hebei University (Grant No. 521000981006), and the Natural Science Foundation of Hebei Province (Grant No. F2021201013).

The authors have no conflicts to disclose.

R.M. and Z.H. contributed equally to this work.

Rui Ma: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Project administration (equal); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Zijun Huang: Data curation (equal); Formal analysis (equal); Methodology (equal); Validation (supporting); Visualization (equal). Wei Ke: Investigation (supporting). Xichen Wang: Investigation (supporting). Peng Hao: Formal analysis (supporting); Funding acquisition (equal); Methodology (supporting); Writing – review & editing (equal). X. Steve Yao: Conceptualization (supporting); Formal analysis (supporting); Funding acquisition (equal); Methodology (supporting); Validation (supporting); Writing – review & editing (equal). Xinlun Cai: Conceptualization (lead); Funding acquisition (lead); Project administration (lead); Resources (lead); Supervision (lead); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding authors upon reasonable request.