In-band full-duplex (IBFD) transmission is an emerging solution for future radio frequency (RF) systems, offering high spectral efficiency by enabling transmitting and receiving signal simultaneously at the same frequency band. However, IBFD systems face a significant challenge in the form of co-site in-band self-interference (SI), where the transmitter interferes with the receiver within the same frequency range. Therefore, the deployment of an RF self-interference cancellation (SIC) technique becomes necessary in IBFD transmission systems. The goal of RF SIC is to eliminate frequency aliasing caused by SI and recover the signal of interest (SOI) with improved quality and fidelity.

Traditionally, RF SIC schemes primarily relied on electrical approaches. However, the limited instantaneous bandwidth and high transmission loss associated with electrical methods have paved the way for a new system known as the IBFD radio-over-fiber (RoF) system^[1,2]. The IBFD RoF system offers several advantages, including large instantaneous bandwidth, broad frequency coverage, low transmission loss, precise wide time-delay tuning, and immunity to electromagnetic interference. In photonic RF SIC approaches, a crucial step is the generation of a phase-inverted replica of the SI signal, which serves as a reference signal for subtracting the received SI. In previous works^[3-5], the utilization of two opposite-biased Mach–Zehnder modulators (MZMs) has been proposed for photonic SIC. Another method described in Ref. [6,7] exploits the two optically inverted-phase sidebands of a phase-modulated signal for achieving photonic SIC. Additionally, balanced photodetection has been employed in Ref. [8,9] as a means of achieving photonic SIC. Realistic applications of RF SIC encompass not only the direct path coupling between the transmitter and receiver but also the internal multipath reflections originating from the RF system^[10]. To effectively mitigate this multipath SI, the RF SIC systems are typically designed with multiple reference signals, allowing them to address multipath reflection components. Several photonic methods have been proposed to obtain multiple phase-inverted replicas of the SI signal to tackle the challenge of multipath SIC. These methods include mode division multiplexing (MDM)^[11] and wavelength division multiplexing (WDM)^[6-8,12]. However, the drawback of these previous approaches is their reliance on discrete components within key optical devices, leading to an expensive system that generates multiple imbalanced reference signals for multipath SIC. With the rapid progress of silicon photonics, substantial endeavors have been focused on harnessing this technology for microwave photonic systems^[7,13-15]. This advancement facilitates the creation of multiple reference signals, offering a cost-effective and higher-performance solution for RF multipath SIC.

In this Letter, we have proposed and experimentally demonstrated a photonic approach for a wideband multipath SIC scheme utilizing silicon photonic techniques. A silicon photonic modulator (SPM) chip is utilized to obtain multiple phase-inverted reference signals. The intrinsic phase relationship between the two optical sidebands in the phase modulators (PMs) of the SPM chip is leveraged, enabling a π phase difference between the multipath SI and reference signal. By employing photonic multi-dimensional multiplexing techniques, such as wavelength and polarization, multiple reference signals can be obtained through the SPM chip and a polarization beam splitter (PBS). Subsequently, a multipath optical tunable delay line (MOTDL) module allows precise manipulation of power attenuation and time delay of the multiple reference signals to satisfy the multipath SIC condition. The experimental results demonstrated that the multipath cancellation depths of 25.53 dB and 23.81 dB are, respectively, achieved for the bandwidths of 500 MHz and 1 GHz by superimposing 2-path reference signals.

The schematic diagram of our proposed wideband multipath SIC method is depicted in Fig. 1. In this commonly used architectural configuration, a single antenna serves both transmission and reception through a three-port circulator. The main sources of multipath SI are circulator leakage and internal antenna reflections, resulting from impedance mismatches between the circulator and the antenna^[10,16,17]. In our approach, two continuous light waves with wavelengths of λ0 and λ1, generated from LD0 and LD1, respectively, are directed into two grating couplers (GCs) of an SPM chip. This SPM chip features two PMs, a MZM, and a dual parallel MZM, as depicted in Fig. 2(a). The packaged external view of the SPM chip is illustrated in Fig. 2(b). Effective management of the SI and reference signal discrepancy is pivotal for optimizing the performance of the RF SIC system, involving the reduction of imbalances in amplitude and inverted phase. The SPM chip plays a crucial role in rectifying this imbalance, thanks to its consistent manufacturing and chip packaging processes employed across various modulators within the SPM chip framework. An RF signal from a transmitter undergoes equal division into two paths. In one path, it is directly routed to one of the PMs within the SPM chip, where it functions as the reference signal. In the other path, the RF signal is directed to the other PM within the SPM chip, where it encounters leakage from the circulator and reflection from the antenna, thereby serving as the multipath SI signal. Subsequently, the modulated optical signals from these individual PMs are combined and routed out of the SPM chip through a GC. This process, illustrated in Fig. 2(b-I), can be mathematically expressed as [E0E1]=[P04ej[ω0t+∑nVs,ncosω0(t−τn)Vπ0/π]P14ej(ω1t+Vrcosω0tVπ1/π)],(1)where P0 and P1 are the output optical powers of the LD0 and LD1, respectively. ω0 and ω1 represent the angular frequencies of the output optical carriers from LD0 and LD1, respectively. Vπ0 and Vπ1 are the half-wave voltages of the integrated PMs, respectively. Vr is the amplitude of the RF reference signal, and Vs,n is the amplitude of the nth RF SI signal. τn is the time delay of the nth RF SI signal. After passing through an optical bandpass filter (OBPF), the successful attainment of optical single sideband (OSSB) modulation is shown in Fig. 2(b-II). It is evident that the two OSSB signals display opposite phases, attributed to the inherent out-of-phase relationship between the two sidebands of the phase-modulated signal. This phenomenon has been explored in other applications, including optoelectronic oscillator, microwave photonic (MWP) filters, MWP receivers, and others^[18-20]. Under the small-signal modulation, the OSSB signal can be conveniently simplified as [E0E1]∝[P04ejω0t{J0(ms,n)−J1(ms,n)e−jω[t+∑n(t−τn)+π2]}P14ejω1t[J0(mr)+J1(mr)ej(ωt+π2)]],(2)where Jn is the nth-order Bessel function of the first kind and mr=πVr/Vπ0 and ms,n=πVs,n/Vπ1 are the modulation indexes of the integrated PMs, respectively. The OSSB signal is demultiplexed into multipath SI and reference signals via WDM. The reference signal is directed to a PBS through a polarization controller (PC) for polarization division multiplexing, illustrated in Fig. 2(b-III). At the PBS, the optically referenced signal is divided into two paths, each exiting through one of the two ports along orthogonal polarization axes, as shown in Figs. 2(b-IV) and 2(b-V). To achieve the necessary time delay (τr,i) and amplitude (αr,i) adjustments, the two orthogonal reference signals are manipulated in the optical domain using a variable optical delay line (VODL) and a variable optical attenuator (VOA). Subsequently, these adjusted reference signal paths are combined via a polarization beam combiner (PBC) as depicted in Fig. 2(b-VI). After that, the reference and multipath SI signals are multiplexed once more using another WDM. The combined optical signals are then directed into a photodetector (PD) for optical–electrical conversion, as shown in Fig. 2(b-VII). The electrical output can be expressed as I∝I0+I1×{∑nVs,ncos[ω(t−τn)]Vπ0−∑i(αr,i)2VrVπ1cos[ω(t−τr,i)]},(3)where I0 and I1 are the amplitudes of the direct current (DC) and fundamental-frequency terms, respectively. In Eq. (3), it can be acquired that the multipath SI signal will be canceled when Vs,n/Vπ0=(αr,i)2/Vπ1 and τn=τr,i. Additional reference signals can be produced by introducing a distinct wavelength from another light source. In such scenarios, the OBPF can be substituted with a periodic OBPF or an optical processor. It is essential to emphasize that both reference and multipath SI signals share a common source. Consequently, the conditions for multipath SIC are inherently met, regardless of the multipath SI signal’s amplitude or frequency.

To verify the feasibility of the proposed multipath SIC method, we have conducted an experimental setup, as shown in Fig. 3. Two tunable narrow linewidth laser diodes (LD0 and LD1, Yenista Optics OSICS) operating at wavelengths of 1550.12 and 1551.72 nm, respectively, with an output power of 14 dBm, were used as the optical carriers. These carriers were coupled into the PMs of an SPM chip via GCs, as shown in Figs. 2(a) and 2(b). The SPM chip, fabricated by Chongqing United Microelectronics Center (CUMEC), possesses a 3-dB bandwidth of 20 GHz. The RF signal, generated by an arbitrary waveform generator (AWG, Keysight M8195A) with a sample rate of 65 GSa/s, was split into two paths using an electrical splitter (ES). One path served as the direct SI signal, while the other path served as the reference signal. The direct SI signal was injected into an RF port of one PM, while the reference signal was applied to the other PM. Along the RF path, the direct SI signal experienced multiple reflections due to impedance mismatches among the electrical devices, simulating circulator leakage and internal antenna reflections. This resulted in varying time delays and power attenuations, ultimately leading to the generation of the multipath SI signal. Moreover, the multipath SI signal could be combined with an SOI generated by another port of the AWG via an electrical combiner (EC). After passing through the PMs, the two modulated optical signals are combined using an internal coupler and then transmitted to an OBPF (EXFO XTM-50) to achieve OSSB modulation and π phase inversion through an output GC. To compensate for coupling and link losses, an erbium-doped fiber amplifier (EDFA, Amonics AEDFA-35-B-FA) was introduced after the OBPF. Subsequently, the reference and multipath SI signals were demultiplexed using a dense WDM (DWDM). The single-path reference signal was divided into two orthogonally polarized optical signals using a PC and a PBS. These two orthogonally polarized reference optical signals were then fed into separate channels of a MOTDL. Within the MOTDL, the time delay and power attenuation of the reference optical signals were adjusted using electronically controlled VOAs and VODLs. The VODLs offered a time delay accuracy of 0.1 ps and a tunable range of 12 ns, while the VOAs provided an attenuation accuracy of 0.1 dB with a tunable range of 30 dB. It is important to note that MOTDL is utilized for easily determining the optimum cancellation depth and addressing system fluctuations or environmental variations. Consequently, achieving tunability in time delay and amplitude becomes necessary in the multipath SIC scheme through the use of MOTDL. The two orthogonally polarized reference signals were subsequently combined using PCs and a PBC. After the combination, the reference signals were multiplexed with the multipath SI signal using another DWDM. The multiplexed optical signals were then injected into a PD (Thorlabs RXM40AF) with a bandwidth of 40 GHz to cancel the multipath SI signal.

The multipath SI and reference optical carriers are modulated by the RF signal with a frequency of 9 GHz. Figure 4 shows the measured optical spectra of the signals after passing through the SPM chip, OBPF, and EDFA. An optical spectrum analyzer (OSA, Yokogawa AQ63670) with a resolution of 0.02 nm is employed to measure the optical spectra. The OBPF selectively filters out the left sideband of the multipath SI optical signal at 1550.12 nm and the right sidebands of the reference optical signal at 1550.72 nm, resulting in the generation of OSSB signals. It is important to note that the phase difference between the optical signals at 1550.12 and 1550.72 nm is π, which enables the achievement of multipath SIC by utilizing multiple reference signals with π phase difference. The proposed scheme operates within a frequency range of 5.5–50 GHz. This is attributed to the OBPF having a filter edge gradient of 800 dB/nm and the optical carrier sources being separated by a wavelength of 0.8 nm.

To investigate the performance of the proposed multipath SIC scheme, a 40 GHz bandwidth vector network analyzer (VNA, Anritsu MS4645B) is employed to measure the cancellation depth curve. A swept single-tone RF signal ranging from 2 to 18 GHz was generated by the VNA and fed into the experimental setup. The resulting cancellation depth curves are captured and presented in Fig. 5. As depicted when employing a 2-path reference signal, the optimal cancellation depth of below 20 dB is achieved within specific frequency ranges: 8.3 to 10 GHz, 11.9 to 12.7 GHz, and 13.2 to 14 GHz. These ranges represent the frequencies where the proposed scheme demonstrates its highest cancellation performance. Therefore, the working frequency region can be selected based on the optimal cancellation depth curve, indicated by the red dashed line in Fig. 5, to ensure effective SIC.

To evaluate the cancellation performance of the proposed method for single-tone transmission, a microwave signal at a carrier frequency of 9 GHz is generated using the AWG and introduced into the experimental system. The electrical spectra are measured and depicted in Fig. 6, showcasing the results for 1-path cancellation (red dashed curve), 2-path cancellation (yellow dotted curve), and without cancellation (blue solid curve). The measurements are obtained using an electrical spectrum analyzer (ESA, Anritsu MS2691A) with a resolution bandwidth (RBW) and a video bandwidth (VBW) both set to 3 MHz. The results show that the cancellation depths of 36.47 and 48.36 dB are achieved for the cases of 1-path and 2-path cancellation, respectively. This demonstrates the effectiveness of the proposed approach in mitigating the adverse effects of multipath SI, with the 2-path cancellation outperforming the 1-path cancellation in achieving higher cancellation depths. Moreover, when comparing our proposed scheme with the utilization of discrete components in Ref. [8], our approach employing the SPM chip showcased superior performance. It provided an improvement of up to 3.97 dB for multipath single-tone signal SIC.

To assess the effectiveness of the multipath SIC scheme for wideband microwave signals, linear frequency modulation (LFM) signals centered at 9 GHz with bandwidths of 500 MHz and 1 GHz are generated using the AWG. The generated LFM signals have an output power of 10 dBm. The measured electrical spectra for three scenarios, namely without cancellation (blue solid curve), 1-path cancellation (red dashed curve), and 2-path cancellation (yellow dotted curve), are presented in Fig. 7. According to the experimental results, the optimal multipath cancellation depths achieved using the proposed scheme are 25.53 and 23.81 dB for transmission bandwidths of 500 MHz and 1 GHz, respectively. Furthermore, our proposed scheme utilizing the SPM chip for multipath wideband SIC exhibits superior performance compared to the use of the discrete scheme^[8]. It provides improvements of up to 0.6 and 1.1 dB over the bandwidths of 500 MHz and 1 GHz, respectively.

To provide a detailed analysis of the wideband multipath SIC, a desired RF signal centered at 9 GHz with a power of −10dBm is selected as an SOI. The SOI is combined with the direct SI signal, which is a 10 dBm LFM signal centered at 9 GHz with bandwidths of 500 MHz and 1 GHz. The measured spectra of the received hybrid signal without and with different-path cancellations are shown in Fig. 8. The results demonstrate the effectiveness of the proposed scheme in recovering the SOI from the strong multipath SI signal when employing 2-path cancellation. In the absence of multipath SIC, the desired RF signal overlaps with the wideband multipath SI signals, creating a challenging scenario where the SOI is difficult to distinguish. This results in low signal-to-interference-plus-noise ratios (SINRs) of 3 and 4 dB for the respective SOIs. However, when incorporating the 1-path reference signal, the wideband multipath SI signals are effectively suppressed, leading to the emergence of the SOI with improved SINRs of 12 and 14 dB, respectively. This signifies a significant enhancement in signal quality. Furthermore, by superimposing the 2-path reference signal, the SOIs experience even greater improvements in SINR, reaching values of 18 and 19 dB, respectively. This demonstrates the efficacy of the proposed approach in mitigating the adverse effects of multipath SI, resulting in clearer and more discernible SOIs. Overall, the results highlight the importance and effectiveness of multipath SIC in improving the SINR and enabling more reliable detection of the desired RF signal in the presence of multipath SI. It is worth noting that the power of SOI exceeds that of the SI in the absence of cancellation, owing to the superposition of constructive interference between SI and SOI.

In conclusion, a wideband multipath SIC scheme with a silicon photonic chip is proposed and experimentally demonstrated. The feasibility of the proposed scheme has been successfully verified through a proof-of-concept experiment employing an SPM chip. The experimental results reveal multipath cancellation depths of 25.53 dB and 23.81 dB for the bandwidths of 500 MHz and 1 GHz, respectively. The proposed method demonstrates the capability of multipath SIC, which holds promising potential for future IBFD transmission systems.