Wideband, Millimeter Wave Domain SI Canceling (>50dB) In-Band Full-Duplex Circulator Receiver

This paper describes a 26-GHz fully integrated In-band Full-duplex Circulator (IBFD) Receiver (RX), which employs passive and active Self-interference Cancellation (SIC) techniques in the mm-wave domain. Coverage of wireless networks at mm-wave frequencies can be enhanced by deploying a large number of base stations economically based on Integrated Access and Backhaul (IAB) relays and repeaters. However, to retain the channel capacity, IAB needs to be implemented using full-duplex schemes that suffer from a strong Transmitter (TX) to RX SI. This SI leakage can significantly impact the receiver sensitivity and increase the baseband/ADC dynamic range requirements. Canceling SI at mm-wave applications is particularly challenging given the high frequency of operation, wide bandwidth, and antenna (ANT) impedance sensitivity to the surroundings. Proposed mm-wave RX with a shared ANT interface based on a circulator with active SI cancelers provide ~53 dB SIC over 400 MHz and ~40 dB SIC over 400 MHz in mm-wave domain to meet the link budget requirements. Proposed architecture achieves SIC by (i) introducing a shared ANT interface based on a hybrid-coupler and a non-reciprocal delay line that provides wideband SIC and additionally creates a SI replica (ii) subsequent active cancellation using SI replica processed with variable gain amplifiers and phase shifters. This system also accommodates SI channel variations due to surroundings. Proposed 26-GHz circulator RX has >100x better SIC at high TX power (>10dBm) levels in comparison to the state-of-the-art and it consumes ~111 mW power. The system is implemented in 45nm SOI CMOS and has an active area of $4.54\, mm^{2}$ . Stand-alone RX NF is ~5.8 dB and TX to ANT insertion loss (IL) is ~3.1 dB. Over-the-air measurements with modulated TX (128 QAM 2.1 Gb/s) and RX (128 QAM 4.2 Gb/s) signals show an EVM of 3.3%.


I. INTRODUCTION
Millimeter-wave frequency bands with wide spectrum allocations promise multi-Gb/s wireless data rates targeted in 5G and beyond. However, higher propagation-loss and shadowing at mm-wave limits coverage [1]- [4]. Therefore, achieving ubiquitous mm-wave coverage requires the deployment of a large number of base stations economically with wireless backhauling [2], [5]- [9]. Potentially, combining base-stations with repeater/relay nodes with spectrum reuse can provide efficient increase in coverage and wireless backhaul, makes network densification economically viable [10]- [13] (Fig. 1).
The associate editor coordinating the review of this manuscript and approving it for publication was Wenjie Feng.
Integrated mm-wave relays/repeaters can be implemented based on half-duplex (HD), or full-duplex (FD) links, as shown in Fig. 2(a). HD links based on time-division duplexing (TDD), frequency-division duplexing (FDD) or spatial/polarization duplexing minimize transmitter (TX) to receiver (RX) self-interference by separating them in time, frequency, or spatial/polarization domains, resulting in trade-offs between network capacity and interference. Inband full-duplex circulator (IBFD) links, provide superior channel capacity through spectrum reuse. However, IBFD links have to handle strong TX to RX self-interference (SI) as shown in Fig. 2(b) and therefore require self-interference cancellation (SIC) >100 dB [11], [12], [14], [15].
SI leakage in IBFD systems from TX to RX path occurs through direct coupling and/or antenna reflections and such  in-band SI cannot be filtered frequency-selective frontend filters. However, SI suppression is necessary to prevent saturation in the RX front-end and reduce required ADC dynamic range. An example of SIC required from a mm-wave RX with TX power, P TX , RX noise figure (NF) of 10 dB over 800 MHz instantaneous bandwidth, BW and assuming SNR min = 20 dB, is given by (1), SIC reqd. > 174 − 10log 10 (BW )−NF + SNR + P TX SIC reqd. > P TX + 95 (1) The high SIC requirement requires distribution of SI suppression across RF, IF and baseband domains along the RX chain [12], [14], [16]- [20]. Digital and baseband cancellation techniques have demonstrated 66 dB suppression of the leakage [20], [21], and hence mm-wave front-end should be able to provide additional ∼40 dB suppression over 800 MHz bandwidth.
Several SIC architectures have been proposed in prior work to enable simultaneous transmit and receive (STAR) at RF and mm-wave. STAR operation with non-reciprocal on-chip circulators using N-path mixers were proposed at RF frequencies in [22] with subsequent works focused on increasing bandwidth, increased power handling and MIMO operation [14], [19], [20], [23]- [25]. Shared-antenna interfaces based on electrical-balance duplexers (EBD) have also been demonstrated at RF in [18], [26]- [30]. However, applying such SIC techniques at mm-wave with shared ANT interface is particularly challenging due to the high operation frequency. Polarization-based cancellation techniques have been demonstrated for mm-wave STAR with multiple feed networks and/or multiple antennas [16], [31], [32].
An FD RX architecture with N-path mixers interfaced to a 90 • -hybrid coupler was proposed in [19]. This architecture ( Fig. 3) had > 40 dB TX to RX baseband cancellation, but does not scale well to higher operating frequency as non-overlapping clocks with a duty cycle of 100%/N are required for the N-path mixer.
A three-port wideband mm-wave IBFD circulator is proposed in [33], [34] based on non-reciprocal delays. These circulators utilize 50% duty cycle quadrature clocks to drive non-reciprocal T-lines or bandpass filters instead of capacitors used in [19], [20], thus ensuring wideband functionality. Non-reciprocity is analytically derived in [33] for all clock signals that are odd sub-harmonics of the mm-wave/RF input signal driving the non-reciprocal delay. This propery enables this architecture to be applied for mm-wave IBFD using subharmonic clock signals. While these IBFD circulators provide low TX-to-ANT IL and wideband SIC, SIC achieved for small signal operation is limited to ∼20 dB at 28 GHz [33] and 40 dB (22 dB at +10 dBm TX SI power) at 60 GHz [34], respectively. Typically, >40 dB SIC at mm-wave is required to meet the link budget at high P TX as shown in 1, and a single stage cancellation is insufficient for targeted SIC.
In this work, we propose a fully integrated full-duplex circulator RX with wideband 'passive' and 'active' SIC techniques in mm-wave domain. SI is first canceled 'passively' with a shared ANT interface based on 90 • hybrid coupler followed by a non-reciprocal transmission line (NTL) delay. Subsequently, 'active' cancellation is achieved by creating a faithful replica of TX and aligning it with the leakage using 37954 VOLUME 10, 2022 programmable pseudo-differential phase shifter and variable gain amplifier (VGA). Multi-stage cancellation facilitates high SI suppression in bands of interest and the system provides integrated SIC of 40 dB over 800 MHz and 50 dB over 400 MHz instantaneous bandwidth for high P TX > 10 dBm. Similar to the approach in [33], the proposed architecture scales well to higher frequencies since the clock frequency required is F RF /N , where F RF is the input RX frequency, and N is an odd integer.
This manuscript expands on [35] to detail the motivation, analysis, system implementation and additional measurements. Prior art is discussed elaborately to place the presented work in perspective of current state-of-theart. Subsequently, detailed analysis is presented to explain the working principle. Additional sub-block level implementation details are also discussed. Further, measurements in the presence of realistic EM environment are added to demonstrate the efficacy of this work. The organization of the manuscript is as follows. Literature and state-of-art IBFD systems based on shared ANT interface along with their challenges at mm-wave frequencies are discussed in Sec. I. The proposed 26-GHz IBFD circulator RX architectural analysis is then presented in Sec. II along with conceptual insights. Sec. III details system implementation in 45nm SOI and the device level architectures of various building blocks of the circulator RX. Measured results are presented in Sec. IV followed by conclusion and future research in Sec. V.

II. PROPOSED MM-WAVE IBFD CIRCULATOR RECEIVER
In this section, the proposed mm-wave SI canceling IBFD circulator-RX capable of providing wideband SIC with high TX power handling, is presented. The section also details approaches that can provide higher SIC utilizing additional avenues for further SI suppression. The proposed architecture is shown in Fig. 4, where, a 'passive' SI canceler is followed by an 'active' one. The following sub-sections present the theoretical analysis of these cancelers.

A. PASSIVE SI CANCELER
Passive SIC is achieved by employing a cascade of two differential quadrature hybrid couplers along with a NTL-based delay proposed in [33]. The working principle of NTL delay and detailed analysis has already been elaborately covered in [33]. In the proposed system, the NTL is driven with a clock frequency of F RF /5 with 50% duty cycle.
Detailed mechanism for supporting simultaneous transmission of the TX through the shared ANT port while receiving the signal incident at ANT port is shown in Fig. 5 and Fig. 6. The TX signal is applied to the isolated port, and ANT is attached to the input port of the hybrid coupler (HC1). Under the assumption of matched input and isolated ports and balanced termination at through and coupled ports, the voltages at N THR1 and N CPL1 can be written as; where, R,TX and R,RX are reflection coefficients for TX and RX signals at balanced ports, N THR1 and N CPL1 . In the following discussion, it will be shown that different R,TX and R,RX can be achieved through different loading at N THR1 and N CPL1 with respect to signals incident at N IN and N ISO . Quadrature relationship between N THR1 and N CPL1 can be established from (2)-(5) as long as ports are balanced, irrespective of the absolute termination magnitude/phase. Hence, the presence of a NTL delay line does not disrupt this quadrature phase relation at the N THR1 and N CPL1 . Furthermore, the NTL delay can be described using the following S-parameters (from [33]): The presence of NTL delay leads to V CPL1,TX combining with jV THR1,TX at N CPL1 and V THR1,TX combines with −jV CPL1,TX at N THR1 (6). From (4-6), the combined signals at N CPL1 and N THR1 observe nulling (V THR1,TX ,EFF = 0 and V CPL1,TX ,EFF = 0) of strong undesired TX signal from PA incident on the isolated port. Nulling of the TX signal creates an effective 'short' across N CPL1 and N THR1 , which leads to reflection of the TX signal to input/ANT port. Consequently, the PA output is transmitted through antenna. Fig. 6 shows the signal-path phase of RX signals incident on the antenna. The RX signal analysis is similar to the TX, but with the important difference that RX signals from ANT port arrive at N THR1 and N CPL1 with phase shifts of − π 2 and 0, respectively. This leads to constructive addition for the RX signals, which is equivalent to an 'open' circuit (in the absence of HC2). The RX signal, is hence available at N THR1 and N CPL1 for further amplification.
In the following, non-idealities will be taken into consideration and the proposed scheme for further SIC to address these non-idealities will be discussed. The discussion of TX leakage cancellation so far has assumed ideal NTL delay, infinite quality factors and ideal switches. In practice, TX nulling is finite and limited by the parasitic capacitance, and resistance, which leads to TX leakage residue at N CPL1 and N THR1 . While the cancellation depth of the leakage depends on the parasitics, TX residue phase ( THR1EFF − CPL1EFF ) is independent of parasitics as long as N CPL1 and N THR1 have balanced loading.
Assuming residual TX leakage at N CPL1 and N THR1 to be δV CPL1,TX and δV THR1,TX respectively, these can be described as; where, 1,TX is the magnitude of the residual TX leakage. Quadrature relationship between δV CPL1,TX and δV THR1,TX , allows the addition of a residual 90 • hybrid coupler (HC2) at N CPL1 and N THR1 to provide further suppression of the VOLUME 10, 2022   residual TX leakage. Further suppression of the residue is possible because RX and residual TX signals have different relative phase shift between N CPL1 and N THR1 equal to + π 2 and − π 2 respectively. The addition of a hybrid coupler does change the 'open' across N CPL1 and N THR1 to 50 for the RX, but quadrature phase relationship still holds. This can also be ascertained from (4)(5) where balanced loads at CPL and THR ports ensure same reflection coefficient ( ) at these ports, leading to quadrature phase relationship irrespective of exact value of .
The residual TX leakage is further canceled by the residual hybrid coupler (HC2) on N CPL2 because of destructive addition of phase-shifted δV CPL1,TX and δV THR1,TX . Additionally, residual TX adds constructively at N THR2 as shown in (9) and (10).
Similar analysis can be performed for the desired RX signal incident on the ANT port. RX and residual TX signals have relative phase shift between N CPL1 and N THR1 equal to + π 2 and − π 2 , respectively, and this leads to RX cancellation at N THR2 and RX constructive addition at N CPL2 . Availability of stronger residual TX opens up another avenue for further SIC in the RF domain considering there is enough isolation between TX signals at Main and Aux nodes (Fig. 6). Isolation of TX residue between N CPL2 and N THR2 is desired if any active auxiliary path is to be used for further cancellation.
The 'passive' SIC scheme discussed above provides additional TX leakage cancellation while ensuring no fundamental power loss for RX signals.

B. ACTIVE SI CANCELER
From Sec. II-A, TX leakage residue cancellation is possible since a stronger TX residue is available at N THR2 . As shown in Fig. 7, N CPL2 connected a Low Noise Amplifier (LNA), and a passive down-conversion mixer followed by a baseband amplifier. Additionally, an auxiliary path through N THR2 phase shifter and auxiliary LNA is proposed. Auxiliary and main paths of RX are combined in the current domain to allow further SIC.
Theoretically, very high suppression is possible through the SI-canceler's 'passive' stage. However, in practice due to finite insertion loss (IL) and phase matching between through and coupled ports, and layout parasitics, SIC is limited, which leads to non-zero V CPL2,TX . Therefore, an auxiliary path for additional cancellation is required.
The auxiliary path is designed to align the amplitude and invert the phase with respect to the main path TX residue using a programmable gm-cell and a wideband RTPS as shown in Fig. 4. Coarse gain programmability in the auxiliary    path is provided by 6-level differential LNA. Phase shifter should be capable of providing 360 • phase shift to ensure cancellation for any possible phase of TX residue. Reflectiontype phase shifter based on [37] can provide 360 • phase-shift  Auxiliary and main paths are current summed at their output to allow better power handling. Residual TX leakage current after summation can be expressed as following: (11) where, I Leak,TX is the residual TX leakage current after addition of main and auxiliary paths, V THR2,TX and V CPL2,TX are the residual TX leakages at N THR2 and N CPL2 respectively due to non-idealities, T is reflection coefficient at through and coupled ports of the RTPS and G m,Main and G m,Aux are transconductances of the main path and aux path LNAs respectively. To ensure a low I Leak,TX , fine gain and phase resolution of the auxiliary path has been provided through the RTPS and variable transconductance.

III. 26-GHz SI CANCELING IBFD CIRCULATOR RX DESIGN IN 45nm SOI CMOS
In this section, the implementation of the proposed mm-wave SI cancellation RX, as shown in Fig. 7, is detailed. The IC is implemented in 45nm RFSOI which provides simulated R on C off of < 200 fs and inductor quality factor >15 due to high substrate resistivity. Block-level circuit implementations of the shared ANT interface and the subsequent RX are discussed in the following:

A. HYBRID COUPLER (HC1) WITH NON-RECIPROCAL DELAY LOAD
The proposed architecture is implemented differentially with an on-chip branch line 90 • hybrid coupler (Fig. 8) designed based on differential 100 and 70 transmission lines. EM simulations were performed to verify mm-wave impedance and wavelength [36]. In this work, T-line based distributed approach is used to design HC1 for increased EM modeling predictability; however lumped components can be used for reducing area. The insertion loss of HC1 when loaded with HC2 is ∼1.45 dB.
Implementation of the switching network for the NTL delay loading the branch-line hybrid coupler is shown in Fig. 7. This design is based on [33], and the NTL delay is clocked at F RF /5 with 50% duty cycle square waves with four quadrature phases. Low-frequency operation assists in two ways; i) reducing the dynamic power consumption of switches, ii) improving the switch performance. Switch resistance is chosen to target TX-ANT IL ∼3 dB. The parasitic capacitance of the switch is absorbed in the T-lines around

C. PSEUDO-DIFFERENTIAL REFLECTION-TYPE PHASE SHIFTER
As described in Sec. II, the residue canceling 90 • hybrid coupler (HC2) drives a phase shifter for further TX leakage suppression. The RTPS is implemented based on the single-ended phase-shifter proposed in [37]. In this work, two stand-alone RTPS are used in a pseudo-differential configuration. To achieve a compact footprint, a transformer-based lumped-element hybrid coupler (Fig. 9) is used to implement the RTPS. Each RTPS is matched to 50 both at input and output. A variable LCL-network is used as the termination, T , at the through and coupled ports with the use of varactors providing variable phase (and resultant variable amplitude). Fig. 10 shows the simulated RTPS amplitude and phase shift (S 21 ) for the RTPS as varactors, C Var,1 and C Var,2 in Fig. 9 are varied, demonstrating phase and amplitude tunability. Since the input and output ports of the RTPS are matched (S 11 in Fig. 10), TX-leakage reflections are not created that can propagate back to main signal path through the residue-canceling hybrid coupler (HC2). Common-source differential LNA with inductive degeneration topology is used for low stand-alone NF of ∼2.45 dB and higher power gain in this work. Fig. 11 shows the VOLUME 10, 2022  input power-matched stand-alone differential LNA architecture. The output of LNA is current summed with the    provides additional amplitude control for canceling leakage using the auxiliary path. Notably, higher TX residue isolation between N THR2 and N CPL2 requires lower auxiliary path gain for active SI cancellation, which leads to lower degradation in noise figure of the RX when auxiliary and main paths are combined. Passive SIC in this design has been optimized to limit the simulated NF degradation due to presence of auxiliary path below 1.5 dB.
A pseudo-differential RTPS drives an input matched programmable auxiliary path LNA. Auxiliary LNA has been designed to always input-match to 50 (single-ended) irrespective of the gain setting. To ensure such matching across all the gain settings, a current steering mechanism is provided Fig. 12. For instance, lower codes steer the excessive current to VDD (1.2V) supply, thus reducing the effective gm to the load. Input matching at auxiliary LNA ensures that the phase shifters witness a matched load at its isolated port. Range of programmability for the auxiliary path LNA is 0 − gm/2, where gm is the transconductance of the main path LNA. RX LNA consumes 27 mW of power, and the auxiliary path consumes 13.8 mW of power.

E. QUADRATURE PASSIVE MIXER AND LO DISTRIBUTION
The current combined output of the main path and auxiliary LNA is directly downconverted with 26 GHz clock driving the passive quadrature mixer, as shown in Fig. 7. Quadrature LO is generated from a single-ended clock using a lumped element 90 • hybrid coupler. Furthermore, the single-ended quadrature signal is converted to differential using baluns. Finally, the LO is buffered through cascoded common source amplifier and gm-cell buffer. LO driver consumes 10.5 mW of power from 1.2 V supply.

F. BASEBAND AMPLIFIER
The passive mixer is loaded with a transconductance stage based on invertors with resistive feedback. Baseband amplifier drives a 100 differential T-line load. Quadrature baseband amplifier consumes 50.4 mW power from 1.2 V supply.

IV. MEASUREMENT RESULTS
The prototype is fabricated in 45nm SOI CMOS process and the micro-photograph of IC is shown in Fig. 13. The active area of the IC is 4.54 mm 2 and the IC is flip-chip packaged onto a RO4350B three-layer printed circuit board for measurements. The measurement setup is shown in Fig. 14 for connectorized CW and modulated signal measurements.
The measured matching at the TX and ANT ports is shown in Fig. 15. Since the input matching is constrained by the matching of the branch-line 90 • coupler, HC1, the distributed implementation leads to wideband matching.
With respect to input at the ANT port, for operation as a receiver, the measured gain from ANT to IF is shown in Fig. 16(a). A peak RX Gain of ∼16 dB is measured for the signal path. Corresponding NF is shown in Fig. 16(b) with a standalone NF of ∼ of 6 dB. When the SIC path is enabled, the NF is degraded as discussed in Sec. III-D. The measured degradation of NF is 2.1 dB. Fig. 16(c) shows measured integrated SIC for two cases; i) with only passive cancellation enabled where the auxiliary LNA output is not summed with the main path ii) with both passive and active cancellations enabled. The passive SIC from NTL at HC1 output and subsequent quadrature combining in HC2 is 22 dB at ∼26.4 GHz. The phase and amplitude programmability in the auxiliary path enables active SIC and the settings can be optimized for notch depth or cancellation bandwidth. As shown in Fig. 16(c), the proposed architecture is configured to achieve an integrated SIC of ∼53 dB over 400 MHz and 40 dB SIC over 800 MHz signal BW at mmwave.
Wideband SIC is demonstrated with measurements shown in Fig. 17. Group delay is measured for, i) the P TX leakage through the main path with auxiliary path powered down ii) leakage through the auxiliary path with the main path powered down. The difference in the group delays between the leakage and cancellation path is shown in Fig. 17. A relatively flat group delay difference of <250 ps demonstrates the wideband transfer function of the cancellation path.
The SIC path is tuned from 25.5 GHz to 28 GHz by varying the varactor bias of the RTPS and using the variable amplitude in the auxiliary path LNA. The tunability in the SIC is shown in Fig. 18 with both passive and active cancelers enabled, where the SIC null is shifted across frequency.
The TX to ANT IL is measured with the SIC enabled and the performance is shown in Fig. 19. TX to ANT IL is primarily due to T-line losses in distributed implementation of HC1 and finite reflection coefficient, R,TX . Measured results show the TX to ANT IL of 3.1 dB. Fig. 20 shows the measured large-signal blocker 1 dB compression point for the RX with respect to input on the TX. For B1dB measurement, the P TX is gradually increased from −10 dBm till ANT to IF witnesses 1 dB compression. Measurements of Fig. 20 demonstrates that power handling capability of wideband FD circulator RX is +11.5 dBm with respect to the TX port.
Connectorized SIC measurements with a modulated TX signal incident on the system are shown in Fig. 21(a). The ANT port is terminated with a 50 , and a signal generator is used as a source for the TX port. A strong 10.25 dBm Peak Envelope Power (PEP) with PAPR of ∼7 dB, modulated (64 QAM) TX signal at 2.4 Gb/s data rate and frequency of 26.41 GHz is applied, and SI suppression is measured for two cases; i) only passive cancellation enabled ( Fig. 21(b)) ii) both passive and active cancellations enabled (Fig. 21(c)). Measurements show SIC of 48.1 dB, demonstrating 100x improvement over state-of-the-art IBFD receivers for large P TX . Notably, large signal SIC performance of the proposed system is comparable to the small signal, demonstrating high power handling.
The robustness of SIC with respect to P TX is demonstrated with Fig. 22 where the power of a modulated 64 QAM TX signal at 2.4 Gb/s is varied and the SIC is measured across different power settings. Notably, RTPS and VGA codes are optimized for highest cancellation around +11 dBm P TX , and are not changed during power sweep. Low variation for a sweep around high P TX demonstrates the robustness of SIC settings to power changes.
SIC performance is also characterized in varying EM environments. Fig. 23(a) shows the setup to measure SIC with a horn antenna attached to the ANT port. In the absence of a mm-wave impedance tuner, variable EM environment is emulated by an external tunable phase shifter and variable attenuator in series with the horn antenna. A global optimization loop is used to find optimum canceler gain and phase coefficients by measuring the TX leakage power and dynamically varying the on-chip RTPS varactor controls. The optimization loop converges to settings that achieve desired objective function of 48-dB integrated SIC across 400 MS/s (64QAM) in <55 cycles as shown in Fig. 23 and measurements show degradation of only about ∼4 dB with metal plate 1 ft from the horn antenna.
Setup and measured results to demonstrate and characterize mm-wave IBFD circulator RX performance for overthe-air (OTA) link is shown in Fig. 25(a)-(c). Two horn antennas are used to excite the ANT port of DUT wirelessly with a desired modulated RX signal at 128 QAM, 600MS/s as shown in Fig. 25(a). Simultaneously, TX signal with 128 QAM and symbol rate of 300 MS/s (limited by instrument) is applied. RX constellation is plotted, and EVM is measured for two cases; i) P TX − P RX =∼0 dB and ii) P TX − P RX =∼ 5 dB. EVM is shown to be 3.3%/4.2% for the two cases respectively in Fig. 25. EVM of 3.3% demonstrates the high power handling and feasibility of the implemented system for full-duplex communication applications such as IAB relays and repeaters. Table 1, compares the proposed architecture with recently published state-of-the-art. This work is the first fully integrated shared antenna interface based IBFD circulator RX at mm-wave. The presented architecture supports high integrated SIC over 400 MHz instantaneous bandwidth for >10 dBm modulated TX signal at 2.4 Gbps. Significant improvement (>100x) in SIC performance at mm-wave frequencies for high P TX is measured. Additionally, modulated large signal measurements demonstrate the robustness of the IBFD circulator RX.

V. CONCLUSION
A 26-GHz in-band full-duplex circulator RX with SIC employing shared ANT interface with 'passive' followed by 'active' cancelers is presented. The feasibility of IBFD circulator RX for mm-wave relays and repeaters is demonstrated through connectorized and wireless measurements. The proposed mm-wave IBFD circulator mitigates the challenge of TX to RX self-interference at high TX powers by suppressing TX leakage at mm-wave frequencies. Wide cancellation bandwidth and low group delay are critical for millimeter wave 5G applications. This system demonstrates programmable SIC bandwidth and integrated TX leakage magnitude. SIC of ∼53 dB over 400 MHz and ∼40 dB over 800 MHz while maintaining a TX to ANT IL ∼3.1 dB is measured. Wireless measurements with modulated RX in the presence of strong modulated TX and varying VSWR demonstrate robustness of the system.

VI. ACKNOWLEDGMENT
The authors acknowledge the support of Dr. Tim Hancock (DARPA), John Bowne (Rohde & Schwarz), and Radu Fetche (Rohde & Schwarz). Dr. Yao Liu and members of High-Speed Integrated Circuits (HSIC) Laboratory, Oregon State University, provided support with equipment and technical inputs.