Wideband image-reject RF channelization based on soliton microcombs (invited paper)

The reception and detection of wideband radio frequency (RF) signals is a critical function in many scientific and technological fields, such as fast detection of vacant time gaps and frequency bands in integrated sensing and communication (ISAC),¹ real-time waveform feature extraction in cognitive radar,^2,3 electronic countermeasures,⁴ and spectrum sensing.^5,6 Many of these applications necessitate the capability to slice a wideband RF signal into multiple narrow sub-bands for parallel reception to match the performance of the back-end digital signal processor (DSP), known as channelization. The traditional channelization is most often carried out by using an analog-to-digital converter (ADC) and subsequent DSP,^7,8 involving signal filtering and decimation in the digital domain. The bandwidth of this channelization scheme is significantly constrained by the sampling rate of the ADC. At present, the sampling rate of commercial ADCs is mostly below 10 GHz, which makes digital channelization unsuitable for RF signals above X-band without the aid of mixers.

The invention of the optical frequency comb (OFC) provides a new mentality for wideband RF channelization.^9–15 Profitting from the plentiful frequency components of the OFC, the multiple spectral copies of the RF signal are generated by modulating the RF signal onto the OFC. Channelization can be achieved using another OFC with a vernier frequency to downconvert the spectral copies or using optical filters with a periodic frequency response to filter the spectral copies. Therefore, the free spectral range (FSR) and the number of comb lines of the OFC determine the working bandwidth and the channel number of channelization, respectively. Electro-optic (E-O) combs have been used primarily for RF channelization because of their simplicity and flexibility.¹⁶ However, a single E-O modulator produces no more than five comb lines,^17–23 which severely limits the channel number. To overcome this, cascaded E-O modulators have been applied to extend the number of comb lines.^24–27 Despite this, the number of channels is still limited to only a dozen or so. In addition, due to the bandwidth limitation of the E-O modulator, the FSR of the E-O comb generally does not exceed 40 GHz. When receiving high-frequency wideband signals is channelized, the overlap of the spectral copies is unavoidable. Therefore, an OFC with a much wider FSR and simultaneously more comb lines is key to improving the performance of channelization.

Recently, some approaches that allow for the creation of OFCs with ultra-wide FSRs spanning optical C–L bands have been applied in channelization, such as those based on shockwave parametric mixers^28–30 and micro-ring resonators (MRRs).^31–35 In Ref. 30, the parametric mixers were used to produce vernier OFCs with 18 comb lines, achieving the channelization with a 1.2-GHz channel bandwidth in the frequency range of 15.5–37.1 GHz. Nevertheless, the use of multiple E-O modulators and pump lasers makes the system structure complex and bulky, which is not the optimum selection for generating the OFCs in RF channelization applications. The emerging technique of MRR-based frequency combs (microcombs) exhibits many advantages (e.g., compact footprint and high coherence) over the conventional technique, which can benefit the RF channelization performance. For example, Tan et al. applied an active MRR and a passive MRR with quasi-matched FSRs to realize RF channelization with 8.08-GHz working bandwidth and 121.4-MHz channel bandwidth.³⁵ The active MRR yielded a microcomb, while the passive MRR acted as a periodic optical filter. Consequently, the spectral copies of wideband RF signals were generated by the microcomb, and the passive MRR filtered the corresponding spectral bins as output channels in the periodic passband. Despite this, to facilitate subsequent digital processing, the channelized output signal generally needs to be an intermediate frequency (IF). This scheme using homodyne detection does not achieve RF signal downconversion. In addition, the crosstalk suppression ratio is limited by the Q factor of the passive MRR, which is only 6.9 dB. Thus, it remains a challenge to achieve RF channelization with a wideband, multi-channel, and a high crosstalk suppression ratio simultaneously.

In this paper, we proposed and demonstrated a wideband image-reject RF channelization scheme. The proposed approach uses dual-MRRs with slight differences in radii to generate the main microcomb and the vernier microcomb, respectively. The main microcomb acts as the signal comb to generate spectral copies of RF signals, and the vernier microcomb acts as the local oscillator (LO) comb. After the selection of dense wavelength division multiplexers (DWDMs), image-reject mixers (IRMs) are used to slice the RF spectrum into a bank of narrow channels by heterodyne detection. In this scheme, the FSRs of both microcombs are near 100 GHz, ensuring that the working bandwidth of channelization reaches 50 GHz (double-sideband modulation) or 100 GHz (single-sideband modulation). The spectra of both microcombs cover the entire C-band, providing sufficient channel numbers. In addition, the IRM based on the balanced Hartley structure is used in this scheme.^36,37 It allows for the downconversion and suppression of undesired mixing spurs of RF signals in the analog domain, including the image interference and frequency doubling spurs, so that the scheme has a higher crosstalk suppression ratio without additional digital signal processing. Through the demonstrated experiments, the RF channelization with 8–37 GHz working bandwidth and 25 channels is carried out. The image rejection ratio is higher than 34 dB (single-tone signal) or 20 dB (1-GHz wideband signal). Therefore, the performance of the proposed scheme is well beyond the existing channelization methods and fulfills the advanced requirements of ISAC, cognitive radar, electronic countermeasures, and other applications.^1–6 

The experimental setup is illustrated in Fig. 3(a). To generate the desired soliton microcombs, two high Q-factor Si₃N₄ MRRs on separate chips are used in the experiment. MRR1 with a designed radius of 227 μm generates the main microcomb with an FSR of 98.55 GHz, while MRR2 with a designed radius of 226 μm generates the vernier microcomb with an FSR of 99.70 GHz. The pump generated by a CW laser (Toptica Photonics CTL 1550) is split into two branches, which are first amplified by erbium-doped fiber amplifiers (EDFAs, Keopsys CEFA-C-PB-HP) to 500 mW, and then, the noise is filtered out by tunable optical filters (TOFs, EXFO XTM-50-S) in each branch. The upper branch pump is injected directly into MRR1, while the lower branch pump is single-sideband frequency-shifted by a dual parallel Mach–Zehnder modulator (DPMZM, EOSPACE IQ-0DKS-25) driven by a voltage-controlled oscillator (VCO, AT-VCO-0510A-20) and then injected into MRR2. Both of the MRRs reach the single soliton state simultaneously by adjusting the pump wavelength, the VCO voltage, and the temperature of two thermo-electric coolers (TECs). After the generation of soliton microcombs, two wavelength tunable fiber Bragg gratings (FBGs) are applied to reduce the pump power. Figure 1(b) shows the spectra of the main microcomb and the vernier microcomb after FBGs, which are measured using an optical spectrometer (Yokogawa AQ6370D). In this experiment, 25 microcomb lines are selected in the dashed region for RF channelization with 25 channels. As shown in Fig. 3(c), the main microcomb and the vernier microcomb walk off from each other, spanning 19.2 nm.

To accurately measure the working bandwidth of the RF channelization, the vernier beats of 25 channels are detected using a high-speed photodetector (Finisar XPDV3120R, 70 GHz) after filtering and coupling two microcombs. Figure 3(d) shows the vernier beats of 25 channels, increasing linearly from 7.40 to 35.05 GHz. Subsequently, the SUT is modulated onto the main microcomb using a Mach–Zehnder modulator (MZM, EOSPACE AZ-DV-65) working at the minimum point to generate spectral copies. After two DWDMs, the spectral copy and the vernier comb line used for the same channel are selected and then fanned out to the IRM, respectively. It should be noted that limited by the number of IRMs available, a pair of TOFs are used in the experiment to mimic the function of the DWDMs, and the performance of each channel is verified in turn. When the SUT is a single tone at 20 GHz, the spectral copy and the vernier comb line for channels 1, 11, 15, and 15 are shown in Fig. 3(e). The BPF with 1–2.2 GHz passband is applied for the IMR in the experiment. Therefore, based on Eq. (4), the SUT will be downconverted to 1.17 GHz and output at the 11th channel.

To verify the ability of wideband RF channelization and the performance of image-rejection, the experiments of RF channelization for two-tone signals are carried out. The frequency components of the two-tone signal are set to $fvern$ + 1.36 GHz and $fvern$ − 1.84 GHz and generated by two RF signal generators (R&S SMA100B), respectively. As a result, both frequency components fall within the passband of the BPF after channelization. For example, the vernier beat of channel 1 $fver1$ is 7.40 GHz, so the frequencies of the SUT are 8.76 GHz (RF signal) and 5.56 GHz (image interference). In the six experiments carried out, the frequencies of the SUTs are shown in the blue and red dots in Fig. 4(a), and the measured image suppression ratios of the different channels are shown in the green dots in Fig. 4(a). Figures 4(b)–4(g) present the channelized output spectra of the SUTs at channels 1, 5, 15, 20, and 25, which are measured using a spectrum analyzer (R&S FSWP). It can be seen that the proposed channelizer achieves the high-performance image interference suppression within 25 channels, and the image suppression ratios are higher than 34 dB. In addition, channel 1 and channel 25 achieve channelized reception of 8.76-GHz frequency component and 36.41-GHz frequency component, respectively, indicating that the working bandwidth of the proposed channelization reaches 8–36 GHz. It should be noted that the main reason for the decrease in the image suppression ratio in Figs. 4(b)–4(g) is the decrease in the signal-to-noise ratio (SNR) of the comb lines. Through reducing the microresonator losses or the dispersion engineering,^38,39 the conversion efficiency of microcombs can be improved, so the working bandwidth and the channel number of the proposed scheme are expected to be further enhanced.

The performance of the channelization of wideband RF signals also is experimentally investigated. Two linear frequency modulation (LFM) signals with different center frequencies are generated as SUTs using an arbitrary waveform generator (AWG, Keysight M8195A). The bandwidth, period, and duty ratio of the two LFM signals are set as 1 GHz, 1 µs, and 0.9, respectively. To mimic the RF signal and the image interference, the center frequencies of the two LFM signals are set as $fvern$ + 1.6 GHz and $fvern$ − 1.6 GHz, respectively. As a result, the channelized output spectrum of both LFM signals is 1.1–1.2 GHz. Figure 4(h) shows the channelized output spectrum of two LFM signals with center frequencies of 9.00 GHz (RF signal) and 5.80 GHz (image interference), respectively, at channel 1. As can be seen, the image suppression ratio of channel 1 is 24.27 dB within the 1-GHz bandwidth. Figure 4(i) shows the comparison of the channelized output time–frequency diagrams with and without image rejection. The left diagram in Fig. 4(i) is the result of direct detection of the spectral copy and the vernier comb line on channel 1 by a PD, while the right one is the result of image-reject mixing using IRMs. It can be seen that the application of IRM effectively suppresses wideband image interference and ensures the crosstalk suppression ratio of the proposed channelization to be good enough. Figures 4(j) and 4(k) show the channelized output spectrum and time–frequency diagram of two LFM signals with center frequencies of 10.16 and 6.96 GHz at channel 2, respectively. The image suppression ratio of channel 2 reaches 20.15 dB. Limited by the performance of the AWG, the experiments of the wideband image-reject RF channelization are carried out on channels 1 and 2.

To better verify the performance of the proposed RF channelizer under strong image interference, the channelization of dual QPSK signals is experimentally carried out. The dual QPSK signals are generated by the AWG with carrier frequencies of 9.26 GHz (QPSK1) and 12.46 GHz (QPSK2), respectively. In addition, the baud rate of the dual QPSK signals is 1 × 10⁷ sym/s. For channel 1, QPSK1 is the RF signal to be received, while QPSK2 is the out-of-band spur. Figure 5(a) shows the channelized output spectra of the dual QPSK signals from channel 1. Since QPSK2 falls outside the BPF passband after downconversion, it almost does not affect the channelized output of QPSK1. As shown in Fig. 5(b), the error vector magnitude (EVM) and constellation of the channelized output from channel 1 are acquired using a spectrum analyzer (Keysight E4440A) with an 80-MHz analysis bandwidth. The measured EVM of the channelized output from channel 1 is 12.55%. For channel 4, the carrier frequencies of downconverted or channelized QPSK1 and QPSK2 are the same, of which both are 1.6 GHz. Therefore, QPSK2 is the RF signal to be received, while QPSK1 is the in-band image interference on channel 4. In contrast to the out-of-band spur on channel 1, the in-band image interference has a more serious impact on the distortion of the RF signal.⁴⁰ The channelized output spectra of the dual QPSK signals from channel 4 are displayed in blue in Fig. 5(c). To investigate the image suppression ratio of channel 4, the AWG is controlled to output only QPSK2, and the channelized output spectra of channel 4 are shown in red in Fig. 5(c). It can be seen that the image suppression ratio of channel 4 for QPSK1 is higher than 25 dB. Figure 5(d) shows the EVM and constellation of the channelized output from channel 4, and the measured EVM is 14.95%, which is slightly higher than that of channel 1.

The linearity of the proposed channelization is evaluated by measuring the spurious-free dynamic range (SFDR) in the experiment. A two-tone RF signal with frequencies of 10.15 and 10.16 GHz is generated by the RF signal generators as the SUT. After channelization, the two-tone signal downconverted to 1.59 and 1.60 GHz is output from channel 2, and the generated components of third order intermodulation distortion (IMD3) are 1.58 and 1.61 GHz. The measured powers of the fundamental signal and IMD3 at different SUT powers are shown in Fig. 6. The noise floor measured at 1-Hz resolution bandwidth (RBW) is −115 dBm/Hz, and the calculated SFDR is 90 dB/Hz^2/3, which can be further improved by reducing the optical link noise and optimizing the signal-to-noise ratio of the used soliton microcombs.

In summary, the proposed scheme achieves RF channelization with a wideband, multi-channel, and a high crosstalk suppression ratio simultaneously. Table I shows the performance comparison between this work and the state-of-the-art photonic RF channelizers. It can be seen that this work achieves a significantly wider working bandwidth and higher channel number compared to channelization schemes based on E-O combs.^17,18,20,25 Meanwhile, this work guarantees an excellent image suppression ratio within the working bandwidth, overcoming the key performance limitation of the RF channelizers based on the shockwave parametric mixers³⁰ and the MRR filtering.³⁵ It should be noted that in the field of coherent signal reception processing,^18,41 the soliton microcombs are better phase-locked to reduce the phase noise of the channelized signal, through phase locked loop and servos.^42,43 In addition, the cost, size, and weight of the system need to be further reduced for practical applications. This challenge is expected to be addressed by using the dual-output IRM to enhance comb efficiency,^17,44 employing the wavelength selective switch (WSS) to improve the flexibility of channel selection,^45,46 or utilizing the compact in-phase and quadrature receiver to reduce the device size.^47,48 Recent developments in programmable multifunctional integrated photonics^49,50 and optoelectronic heterogeneous integration^51,52 enable the integration of microwave photonic devices, such as 90° optical hybrids⁵³ and BPDs.⁵⁴ In particular, the iteration of the integrated microwave photonic filter subverts the bulkly fiber-based optical filter,⁵⁵ which lays a foundation for the miniaturization of the proposed RF channelizer on a chip.

In conclusion, the wideband image-reject RF channelization has been implemented using integrated dual-soliton microcombs. In the demonstration experiment, two soliton microcombs with the FSRs of 98.55 and 99.70 GHz spanning the optical C-band are generated as the main and vernier microcombs by the designed MRRs. Compared with the typical photonic RF channelization, the ultra-high FSRs ensure that the spectral copies of wideband RF signals do not overlap each other, enabling the working bandwidth of the proposed RF channelization to reach 8–37 GHz. In addition, the multiple comb lines of the soliton microcombs achieve the RF channelization with 25 channels in the experiment. The performance of the proposed RF channelization is investigated using different SUTs, such as two-tone signals, wideband LFM signals, and dual QPSK signals. The experimental results show that the image suppression ratio is higher than 34 dB for single-tone signals and 20 dB for 1-GHz wideband signals. When receiving the QPSK signals with image frequency interferences, the EVM of the channelized output signal still reaches 14.95%. The SFDR is also measured to be 90 dB/Hz^2/3 at last. Therefore, this work represents a significant advancement in the field of wideband RF channelization, featuring multi-channel, excellent crosstalk suppression, and a compact footprint. The proposed scheme not only paves the way for integrated RF channelizers but also reaffirms the revolutionary impact of microcomb technology on building a fully integrated photonic RF system.

This work was supported in part by the National Key Research and Development Program of China under Grant No. 2022YFB2802700; in part by the Natural Science Foundation of Jiangsu Province under Grant Nos. BK20220887 and BK20220076; in part by the National Natural Science Foundation of China under Grant Nos. 62205145, 62271249, and 61971222; and in part by the Nanjing University of Aeronautics and Astronautics under Grant No. 3082022NS2022044.

The authors have no conflicts to disclose.

J.D. and Y.W. contributed equally to this work.

Jiewen Ding: Conceptualization (equal); Investigation (equal); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Yifan Wu: Conceptualization (equal); Formal analysis (equal); Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Huashan Yang: Methodology (equal). Chao Zhang: Methodology (equal). Yifei Zhang: Methodology (equal). Jijun He: Supervision (equal); Writing – review & editing (equal). Dan Zhu: Supervision (equal); Writing – review & editing (equal). Shilong Pan: Supervision (equal); Writing – review & editing (equal).

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