A Wideband If Receiver Chip for Flexibly Scalable mmWave Subarray Combining and Interference Rejection

Large-scale multibeam phased array systems suffer from interbeam interference (IBI) that should be canceled either in the analog or digital domain. In wideband systems such as fifth generation (5G), interference rejection over a wide bandwidth is challenging to achieve, not only due to nonidealities of the receiver chain but also due to the properties of the radio channel. This article presents a scalable IBI cancellation (IBIC) scheme at intermediate frequency (IF) using an IF receiver (IF-RX) chip. The IF-RX provides the flexibility of not just interference rejection between the subarrays but also wideband signal combining over multiple subarrays. It also provides wideband filtering before the analog-to-digital converter (ADC) to support 5G channel bandwidths of up to 800 MHz, high linearity, and low noise figure. A calibration method is proposed to find the cancellation coefficients for the IF-RX in measurements. Furthermore, a simplified over-the-air (OTA) IBIC model for analyzing rejection bandwidth limitations is presented. Interference rejection performance is demonstrated through the OTA measurements using 5G new radio (5G NR) signals. In the OTA measurements, 34–37-dB rejection was achieved for 50–100-MHz signals, while error vector magnitude (EVM) requirements of the 5G standards were met with good margins. Finally, the interference rejection over $4 \times 100$ MHz carrier aggregated 5G NR waveform was demonstrated.

Abstract-Large-scale multibeam phased array systems suffer from interbeam interference (IBI) that should be canceled either in the analog or digital domain.In wideband systems such as fifth generation (5G), interference rejection over a wide bandwidth is challenging to achieve, not only due to nonidealities of the receiver chain but also due to the properties of the radio channel.This article presents a scalable IBI cancellation (IBIC) scheme at intermediate frequency (IF) using an IF receiver (IF-RX) chip.The IF-RX provides the flexibility of not just interference rejection between the subarrays but also wideband signal combining over multiple subarrays.It also provides wideband filtering before the analog-to-digital converter (ADC) to support 5G channel bandwidths of up to 800 MHz, high linearity, and low noise figure.A calibration method is proposed to find the cancellation coefficients for the IF-RX in measurements.Furthermore, a simplified over-the-air (OTA) IBIC model for analyzing rejection bandwidth limitations is presented.Interference rejection performance is demonstrated through the OTA measurements using 5G new radio (5G NR) signals.In the OTA measurements, 34-37-dB rejection was achieved for 50-100-MHz signals, while error vector magnitude (EVM) requirements of the 5G standards were met with good margins.Finally, the interference rejection over 4 × 100 MHz carrier aggregated 5G NR waveform was demonstrated.Index Terms-Beamforming, complex multiplication, interbeam interference cancellation (IBIC), phase shifter, phased array.

I. INTRODUCTION
F IFTH-generation (5G) communication utilizes large-scale phased arrays to enhance the signal-to-noise ratio (SNR), link budget, and spatial selectivity at millimeter-wave (mmWave) frequencies [1].Large-scale phased array systems utilize a subarray approach to provide flexibility and scalability in system design based on specific link budget requirements.These subarrays can then be used in a variety of ways: signal contributions can be summed to increase beamforming gain [2], [3], [4], [5], [6] or used in a multibeam phased array system by assigning each subarray to a single user [7].Each subarray user beam may experience interbeam interference (IBI).In particular, when an unwanted beam falls close to the main-lobe or sidelobe maximum, interference may exceed the desired signal before digitization, resulting in high dynamic range analog-to-digital converter (ADC) requirements [8], [9].As a result, it is critical to provide IBI cancellation (IBIC) in the analog domain to improve the signal-to-interferenceplus-noise ratio (SINR) before the ADC to relax dynamic range and thus power consumption as motivated in [10] and demonstrated in this article.
Recent works [3], [4], [11] focus on combining signals from multiple subarrays at intermediate frequency (IF) and exploit orthogonal polarization to increase the channel capacity by creating independent user beams.However, in the presence of multiple users within the same polarization, interference cancellation is required to achieve a decent SINR and data rate for all beams under various user conditions.Active interference cancellation can be realized, for example, by creating the null/notch in the direction of an interferer in the spatial domain by a feedforward cancellation path or by using zero-forcing techniques in the digital domain after the ADC.In addition, any combination of different digital and analog techniques can be used.To maximize the capacity and frequency reuse, this should be done over the entire bandwidth of the whole receiver.
For 5G new radio (5G NR) sub-6-GHz or mmWave systems having radio frequency (RF) bandwidths of hundreds of MHz, wide cancellation bandwidth is required.In [12], adaptive null steering is achieved in an mmWave multi-in multi-out (MIMO) receiver by directly employing mmWave transconductor currents for phase shifting.However, null steering in one direction requires antenna-specific amplitude control and it may potentially increase the interference in other directions.Alternatively, Huang and Wang [13] proposed beamforming and multistage interference rejection at an IF of 2-5 GHz.Narrowband phase shifters are used for beamforming and the cancellation paths employ tunable passive delay lines and variable-gain amplifiers (VGAs) to achieve wide rejection bandwidth and compensation of path differences.However, the implementation of passives with large area penalty makes it less attractive for low-gigahertz frequencies.In [14], a twostage fully connected subarray architecture that performs zero forcing in the digital domain is demonstrated.However, this increases the number of required ADCs and increases their dynamic range requirements.On the other hand, the cancellation at RF, implemented at the low noise amplifier (LNA) output in [15] at 10 GHz, requires band-limiting quadrature hybrids, degrading the receiver noise performance and providing narrowband interference cancellation.
In MIMO receivers operating in the sub-6-GHz bands, recent papers have focused on spatial notch filtering approaches [16], [17], [18], [19], [20].Two stages of the interference cancellation are proposed in [16] based on a spatial notch filter and feedforward cancellation path realized in the voltage domain at baseband (BB), which results in limited bandwidth of the notch rejection and linearity.An improvement of the notch bandwidth and linearity is demonstrated in [17] by exploiting current-mode passive mixer avoiding large voltage swing at the mixer output to improve the linearity.Golabighezelahmad [18] demonstrates wideband interference rejection and high linearity by the slicing of RF transconductors using directly the currents to provide phase shift by steering the current with static reconfiguration switches to low-ohmic, wideband transimpedance amplifier (TIA) nodes.Alaei et al. [19] demonstrated adaptive interference rejection both in analog and digital domains and included over-the-air (OTA) measurements using vector modulator (VM) of [18] for RF frequencies up to 6-GHz and narrowband (1 MHz) signals.
This article presents a broadband mmWave subarray-based architecture concept for spatial filtering and IBIC to improve the SINR between independent beams using multiple wideband IF receiver (IF-RX) chips [21].Compared with [21], this work further investigates the performance of the proposed architecture concept for wideband IBIC using the IF-RX in the presence of circuit matching and OTA channel nonidealities.This article also analyzes the limitations of the interference rejection bandwidth (BW Canc ) caused by nonidealities in the circuit and the OTA channel.Furthermore, we also propose a calibration method to find the cancellation weights based on the power and error vector magnitude (EVM) measurements.
System considerations are discussed in Section II.The IF-RX chip architecture and circuit details are shown in Section III, proposing flexible summation of parallel IF-RX paths either for improved directivity or interference rejection.Section IV discusses the factors affecting the rejection precision, concentrating on the delay mismatch impact due to the OTA channel between two antenna signals.Compared with [21], Section V shows additional measurements of IF-RX, and Section VI illustrates the rejection performance up to 400 MHz in OTA measurements with modulated orthogonal frequency-division multiplexing (OFDM) signals used in the 5G NR standard.The performance is compared against the state-of-the-art.Finally, Section VII concludes the work.

II. SYSTEM CONSIDERATIONS
A common approach in a multibeam system is to have a dedicated subarray for each user that can be physically separated.Each subarray performs analog beamforming by directing the beams toward one specific user that transmits one data stream, spatially filtering out-of-beam interferers, and downconverting the mmWave signal to IF or BB, as shown in Fig. 1(a).While every subarray provides spatial filtering, each beam may still experience relatively high sidelobe levels (SLLs), resulting in limited spatial filtering and thus SINR from beams coming from other directions.Furthermore, the received signals may have different power levels, especially in multiuser scenarios, leading to even higher IBI.An additional interference rejection scheme is then required to improve the SINR for all the received beams.
In this article, IBIC is performed in the analog domain at IF to receive multiple independent data streams while improving the SINR of each beam using the IF-RX chip of [21].The architecture of the IF-RX chip, shown in Fig. 2, provides the following functions: 1) vector modulation; 2) Cartesian combining of two input signals; 3) zero-IF downconversion; and 4) low-pass filtering for channel selection.Based on this IF-RX chip, two architectures are proposed for IBIC, as shown in Fig. 1(b) and (c), considering a two-user scenario.Fig. 1(b) shows the IBIC scheme with mmWave subarrays, which applies the complex beamforming weights at mmWave before downconversion to IF.The complex weighted signals of each antenna element in each mmWave subarray are summed at mmWave and downconverted to IF, providing single voltage output V C (sum of signals after complex weighting).Here, the IF-RX chip works as a VM to realize IBIC at IF. Fig. 1(c) shows IBIC for a Cartesian combining architecture.Here, each antenna element splits into two paths named real and imaginary path, provides scalar-only beamforming weights with VGAs named Ar 1-n and Ai 1-n (amplitude weights for real and imaginary) at mmWave [22].The scalar-only weighted signals of each path are summed at mmWave and downconverted to IF, providing V Ar and V Ai voltage outputs, respectively.Cartesian combining (V Ar + jV Ai ) is implemented in the IF-RX chip that implements the 90 • phase shifting, which can be done more accurately at the lower IF-frequency.The I/Q generation at low-IF can be done by using digital dividers, which can have a large tunning range, driving I/Q mixers to realize 90 • phase shift.This avoids the generation of 90 • phase shift for each antenna element over a broad range of mmWave frequencies [22], [23].It relaxes the mmWave subarray design that is already challenging and now only needs mixers and VGAs, not phase shifters.On the other hand, the potential challenge for the architecture of Fig. 1(c) is high mixer linearity because the quadrature combining for beamforming is realized after mixing, as a result exposing mixers to potentially stronger interferers.
In both architectures shown in Fig. 1(b) and (c), the mmWave subarray module optimizes the SNR of the receiver, realize the spatial filtering of mutual interference between user1 and user2, and reduces the magnitude of possible multipath reflections from different directions.Thus, prior to the IBIC, mmWave spatial filtering offers wideband signal rejection.The IBIC scheme requires amplitude matching and phase rotation such that the user2 signal is canceled in the user1 path and vice versa.To this end, each mmWave subarray IF output is split into an IF-RX main path (M) and a combining or cancellation path (C), as shown in Fig. 1(b) and (c).The IF-RX M1 and M2 paths receive user1 and user2 streams and provide channel selection filtering for wideband signals.IF-RX C1 and C2 paths can be used for combining signals of multiple subarrays implemented on separate modules/ICs into one BB output or as interference cancellation paths, as studied in this article using the architecture shown in Fig. 1(b).
The dual-conversion IBIC scheme exploits the constant-gm VM concept [24], previously used for interference rejection in zero-IF MIMO receivers [18].Although similar in circuit topology, the system and circuit requirements are quite different.In a low-GHz array with a limited number of antenna elements (4 × 4 MIMO) [18], [19], it is feasible to realize programmable complex weights for each antenna element.This becomes unpractical for mmWave arrays that require much more antenna elements.In this work, each IF-RX chip accesses already weighted and summed mmWave signals to reduce the number of IF signals to two per mmWave subarray (V Ar and V Ai ).The number of IF-RX chips depends on the number of data streams (Ns) and not the number of antennas.To enable IBIC, Ns data streams require one IF-RX chip in the main path and (Ns-1) cancellation paths at the output of each mmWave subarray.In addition, each cancellation path uses attenuators (ATT) in front of the VM to broaden the achievable dynamic range, whereas work [18] relies on the VM only for amplitude and phase matching when synthesizing beam patterns with arbitrary nulls.Thus, it loses phase resolution when main and cancellation path signal levels have a large amplitude difference.This work uses ATTs to do the amplitude matching and operate the VM in the maximum phase resolution, thus providing better rejection.It allows the VM to achieve 26-dB rejection with 5-bit phase quantization [25].
In scenarios with less interference, some cancellation paths can be redundant.They can still be useful to improve the signal quality by combining multiple subarrays into one BB output when subarray beams point to the same user.The rest can be utilized for cancellation.This approach provides the flexibility of scaling and selecting the mmWave subarrays and the number of data streams.To be effective, the IF-RX chip for mmWave 5G NR system should feature: 1) large BB bandwidth to process wideband signals available at mmWave; 2) moderate noise figure (NF) after subarray combining; 3) good linearity to handle the combined signal dynamics; and 4) the capability of interference rejection over a wide bandwidth.

III. IF-RX MODULE
The constant-gm VM was originally proposed with capacitive voltage output in [24] and with TIA output designed for zero-IF MIMO in [18].Here, we modify it to perform Cartesian combining (90 • phase shift) and a VM function compatible with the mmWave subarrays.Hence, unlike [18] where each antenna provides a single-wire (real) input signal to the VM, here we need to support a two-wire input signal to realize complex signal, V Ar + jV Ai , as shown in Fig. 1(c).The 90 • phase shift to realize Cartesian combining is implemented in IF-RX exploiting quadrature mixing, as shown in Fig. 2.

A. VM Function
The architecture of the IF-RX chip is shown in Fig. 2. It uses a low noise transconductance amplifier (LNTA) for V -I conversion and downconverts the IF between 2 and 4 GHz to BB.The VM function is realized by dividing the IF-RX into parallel slices (amplitude weighing) while varying the phase with a selection block implemented by static phase selection switches in each slice ("SW" in Fig. 2) [24].Instead of exploiting charge sharing on a capacitor averaging the Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
output as in [24], Cartesian combining is implemented by current summing into a virtual ground node provided by a TIA, as shown in Fig. 2.
The Cartesian combining and VM functionality can be understood by calculating signals at different nodes of the IF-RX.Consider that the voltage input signal is represented in terms of its complex envelope x(t), and then, the real and imaginary paths' sum terms can be given as Those are split into binary weighted slices, and each unit slice contains constant gm stage and double-balanced passive quadrature mixing.The gm-stage converts IF input voltage to current, i.e., I Ar = gmV Ar and I Ai = gmV Ai .Assuming low impedance at the TIA input node, this current is downconverted to BB by the four-phase double-balanced passive mixer.The BB current I bbi (t) at node I slice can be written as and BB current where M is the number of slices, ω bb = ω H −ω LO , and 2 √ 2/π is the magnitude of current in BB.The phase selection block with reconfiguration switches introduces a static phase rotation function.Each slice operates as a 2-bit VM, which has output phases of 0 • , 90 • , 180 • , or 270 • .By combining multiple slices in a binary weighting scheme, a higher number of bits in the VM constellation is accomplished (see Fig. 2).The TIA sums up the BB current from all VM slices at I TIA and Q TIA and is converted to voltage output by the feedback resistor R FB .This combined output voltage in BB can be expressed as where m, n, o, and p are selecting the phase of binary weighted slices.Equation (3) shows that all the weights are always used in each VM constellation point.All VM constellation points are generated by a combination of binary weighted slices with different phases.Equation ( 3) also shows that each slice in IF-RX provides the 90 • combining of V Ar and V Ai with additional weights for combining/cancellation.For complex weighted, V C summed input from the mmWave subarray, as shown in Fig. 1(b).The Cartesian combining function of the IF-RX chip is not required.In that case, the IF-RX chip can be used with input V Ar alone.By terminating V Ai , Ai in (3) will be zero and IF-RX will only operate as VM for signal combining of mmWave subarrays to improve SNR and/or IBIC.The IF-RX chip processes V Ar and V Ai input signals, e.g., when it would be integrated with [23], both inputs are needed.For routing simplicity, we used the architecture of Fig. 1(b) here to demonstrate and characterize IBIC.Hence, we only need one input.

B. IF-RX Circuit Design
The amplitude and phase constellation points of the VM depend on the number of slices.For M slices, the constellation consists of M + 1 by M + 1 points and contains 3M − 1 amplitude points on the largest circle that can be drawn inside the boundaries of the constellation.The implementation of the IF-RX is realized by dividing the input into 15 binary weighted, parallel slices where each slice acts as an independent vector component in the VM.The transistor-level implementation of a single slice is shown in Fig. 3.Each slice consists of a differential transconductor (gm), a doublebalanced quadrature passive mixer for each differential V Ar and V Ai inputs and phase selection block at BB that adds current contributions of slices at the virtual ground node of the TIA.The static switch configuration is designed such that the current from each transcondutor always contributes to some output [18].The TIA provides both BB I -V conversion and BB low-pass filtering.An external reference clock at twice the local oscillator (LO) frequency is used to generate the fourphase 25% duty cycle clock by an on-chip current-mode logic divider and AND gate logic.The resulting clock is distributed across the chip to drive the mixer switches.

C. Transconductor Stage
Two inverters serve as a pseudo-differential transconductor pair (gm) acting as LNTA and provide the input matching at the IF port (see the center part of Fig. 3).In the inverter, both nMOS and pMOS contribute to the total transconductance and resistive feedback RF provides low NF and low impedance match due to the Miller effect [24].The gm stage is dc-coupled with mixer switches to avoid the large ac-coupling capacitor in each slice.To avoid the dc operating point mismatch, a common-mode feedback (CMFB) stage is added to each differential pair and controlled independently through the current DAC.The pMOS transistors M 3 and M 5 are controlled by CMFB, and they feed the current at the output node of the gm stage depending on the common mode and bias input and output of the gm stage.Single slice gm stage devices use 7-µ/40-nm transistors, including a feedback transistor with 3.5-k feedback resistor R F .The single slice provides a total gm of 8 mS.Each slice provides an input impedance of 800 .By combining 15 parallel slices, the input impedance is ≈50 .

D. Quadrature Downconversion
The Cartesian combining of the inputs V Ar and V Ai and the downconversion is realized with double-balanced quadrature passive mixer, as shown in Fig. 3(a).The mixer consumes  no static power and has a theoretical minimum conversion loss of 0.9 dB for 25% duty cycle with an additional 0.5-dB loss of Cartesian combining.The gate terminals (LO bias ), sources (biased from the gm side), and drains (biased from the BB transimpedance amplifier (BB-TIA) side) of the mixer switches are biased to half of the supply and shift the dc level of the LO locally to fully switch ON and OFF the mixer switches [Fig.3(a)].The I/Q signals generated in the mixer pair are further processed by the phase selection block.The switches are driven by static control signals and allow to pick one of the four mixer phases, which effectively rotate the phase by 0 • , 90 • , 180 • , or 270 • [Fig.3(c)].

E. BB-TIA Circuit
The inverter-based BB-TIA is implemented as in Fig. 4 to provide a better tradeoff between power and noise performance compared to a single-stage Op-amp-based BB-TIA [26].The BB-TIA is self-biased to approximately half of the VDD and has CMFB at the output to reduce the common-mode gain and even-order distortion.The TIA bandwidth depends on the R FB , C FB , input node capacitance, and output node load impedance.The pole frequency of the TIA voltage transfer function is For a fully integrated system, TIAs would typically directly drive an ADC or a buffer before the ADC.In our measurements, the TIA directly drives single-ended R L = 50 and C out = 4 pF.The core of the TIA consists of two inverters, in which each of them drives a single-ended 50-load.High gm TIA = 100 mS is used in the TIA stage, increasing the TIA's current consumption.Low load impedance reduces the gain of the TIA but extends the bandwidth.

A. Interference Rejection
Interference rejection of narrowband signals can be done by summing two signals with opposite phase.It has been applied for many purposes, including reduction of nonlinear distortion [27], to avoid LNA saturation in in-band full duplex transceivers [28], and also for spatial filtering [16], [17], [18], [19], [29].However, achieving sufficient amplitude and phase accuracy for cancellation purposes in circuit solutions is challenging over a wide bandwidth.Decent rejection has been demonstrated only for a rather narrowband modulation.In the case of 5G NR having RF bandwidths of several hundreds of megahertz, this will become a major issue for any broadband cancellation scheme.Furthermore, even small time delays in a radio channel and differences in circuit realizations will impact the cancellation bandwidth (BW Canc ).
In the case of beam-steered mmWave subarrays, the wideband rejection brings in one more challenge when two subarrays are receiving from different directions while still interfering with each other through sidelobes or even inside the main lobe.Then, the canceller [IF-RX C1 or IF-RX C2 in Fig. 1(b)] needs weights tuned to the interfering signal direction.Thus, the cancellation weights in IF-RX VM can differ from those in the mmWave subarrays.Hence, a calibration method for finding the most appropriate coefficients for the IBIC is needed.

B. Aspects of Wideband Rejection
Consider the simplified OTA model of the proposed architecture for IBIC in Fig. 5 to illustrate the limitations of the wideband interference rejection.The interference observed in A1 from TX2 is simulated by placing TX2 at an arbitrary angle Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.2) OTA Channel Delay Effects: In the OTA channel time delays, τ 1 and τ 2 are slightly different due to the angle and location of the interferer (TX2) with respect to the physical positions of the receiver elements and arrays.It means that the cancellation and main paths are summed together with a delay difference.As shown in Fig. 6(b) and (c), BW Canc will be further narrowed compared to frequency-flat phase shifter due to the fact that the slopes of the phase responses are different.When branch mismatch errors are the worst, the cancellation BW Canc will be more sensitive to small delays.It imposes a limit on how much delay the system can tolerate for a particular BW Canc .Third, multiple propagation paths may exist even in the line-of-sight (LOS) channels, and they usually have lower amplitude but much longer delay (τ 3 ) than the main propagation paths (τ 1 and τ 2 ).Even small reflections will cause ripples in signals seen in the main and cancellation paths, limiting the cancellation further.In a simple two-path model, the ripple frequency depends on the delay in the reflection (this effect is further discussed in Section VI).

C. Power and EVM-Based Calibration Method
In order to cancel the interference, the cancellation coefficients must be first found.This can be done in many ways in the digital and analog domains depending on the hardware implementation.For example, Cao and Zhou [30] and El Sayed et al. [31] implemented analog circuitry using the least mean square (LMS) method to train the cancellation, while Nagulu et al. [32] optimized gains and delays using the digital iterative feedback loop algorithm.In [21], the cancellation coefficients were found by using exhaustive search.In this article, the main objective is to reach cancellation performance over the entire signal bandwidth.To this end, we apply powerbased and EVM-based calibration method to calibrate the VMs for good cancellation performance.Hence, the calibration method has two steps.In the first part, the aim is to find the phase shifter value that minimizes the power of the sum signal containing the interference.The second part is simple fine-tuning using EVM measurements, mainly to verify the cancellation performance.
Let the signal observed at the A1 output be s 1 = x 1 + a 12 e jφ 12 x 2 and signal at A2 output s 2 = a 21 e jφ 21 x 1 + x 2 , where x 1 and x 2 are the received signals transmitted from the TX1 and TX2, respectively, and a nm and φ nm are the relative amplitude and phase of signal n observed at subarray m, respectively.After cancellation, the aim is to see only x 1 in the A1 output and only x 2 at the A2 output.Let us first calibrate the cancellation to cancel x 2 from the A1 output.After the RF beamforming, it is expected that a 12 < 1 and a 21 < 1, i.e., the signal-to-interference ratio (SIR) is positive in decibels at both array outputs.Let us denote the VM phase value applied for the cross-coupled path from A2 to A1 as φ ps .By summing the phase-shifted version of the A2 output to the A1 output, we get s 1,φ ps = s 1 + a 12 e jφ ps s 2 = (1 + a 12 a 21 e j (φ 21 +φ ps ) ) ≈1,a 12 a 21 ≪1 x 1 + (a 12 e jφ 12 + a 12 e jφ ps )x 2 ≈ x 1 + a 12 (e jφ 12 + e jφ ps )x 2 = x 1 + a 12 e jφ 12 (1 + e j (φ ps −φ 12 ) )x 2 = x 1 + 2a 12 e j 1 2 (φ ps +φ 12 ) cos 1 2 Now, assuming that x 1 and x 2 are independent signals, the power P 1,φ ps of the sum signal at the A1 output with phase shifter value φ ps can be written as The result has maximum (= P 1 + 4a 2 12 P 2 ) when φ ps = φ 12 and minimum (= P 1 ) when φ ps − φ 12 = 180 • , which gives the cancellation based on (4).Based on (5), finding the cancellation phase turns into a problem of finding the initial phase of the cosine in (5) with a minimum number of used phase shifter values φ ps .The phase shift can be measured by sampling it four times over the period of the cosine.In general, this means that we need to measure with four different phase shifter values to get the desired cancellation coefficient.Hence, we measure with φ ps = {0 • , 90 • , 180 • , 270 • } and note the points as {P 1,0 , P 1,90 , P 1,180 , P 1,270 }.Based on these, the real and imaginary parts whose relation describes the required phase shift of the cancellation path can be derived as Re(e jφ canc ) = 1 2 (P 1,270 − P 1,90 ).
The phase that minimizes (5) can be calculated using arctan(Im/Re).Note that the correct phase can also be realized by applying the real and imaginary values given in ( 6) and ( 7) directly to the VM after normalizing the phasor absolute value to one.More points can also be measured to improve the estimation accuracy.A similar analysis can be made for canceling A1 output from A2 output.Note that the described process aims first for calculating the phase of the cancellation coefficient.Hence, even if the applied amplitude coefficient a 12 would not be accurate, the sum signal observed over the phase shifter states has still the same minimum.Hence, once the minimum is found, the system can retune the amplitude to improve the performance.Initial guess of a 12 can be estimated, for example, based on SIR (linear scale) observed at A1 output without cancellation as a 12 = 1/SIR 1 .This can be derived, for example, based on received EVM or SNR or even based on the known calibrated beam shape relation with respect to the steering directions.
Choosing a phase value to minimize the power of the interference results in zero-forcing coefficients.However, this does not necessarily mean that the signal quality is maximized due to the fact that the interference nulling may have a slight impact also on the useful signal.This is the case especially when the interference is rather strong.The second stage of the calibration method takes a fixed amount of EVM measurements with different phase shifter values from both sides of the estimated power minimum to tune the calibrated coefficient for minimizing the EVM.The two-stage calibration method is demonstrated by OTA measurements in Section VI-B1.

A. General Setup
The IF-RX was realized on-chip in a 45-nm PDSOI CMOS process.The chip has an active area of 0.28 mm 2 , including clock generation.The circuit consumes 104 mW of power from a 1-V supply and 10 mW from a second 1.5-V supply for buffers to drive static switches.Fig. 7(a) shows a chip photograph with the placement of various receiver blocks with their power consumption.The IF-RX chip measurements are performed using external 1:2 impedance baluns at the RF, BB, and LO ports of the printed circuit board (PCB) shown in Fig. 7(b).
Conversion gain and linearity measurements were performed using a four-port vector network analyzer (VNA).An external splitter followed by passive balun on PCB is used to drive differential V Ar and V Ai RF inputs.The BB I and Q outputs are measured through passive baluns on PCB, as shown in Fig. 8    The setup described in Fig. 8 is used to measure NF with a UXA N9040B signal analyzer.Only the V Ar input is used and V Ai is terminated.At a fixed 50-MHz BB, a double sideband NF (NF DSB,eq ) of 5.5-8.5 dB is measured across the LO frequency and 5.5-9 dB is measured for variable BB at the fixed RF band, as shown in Fig. 10(a) and (b), respectively.The measurement shows a minimum of 5.5-dB NF DSB .Because the IF-RX chip is placed after the mmWave module with LNA and beamforming in the system, its noise contribution to overall system noise is minimal.The measured third-order intercept point (IIP3) both for in-band and outof-band (OoB) is shown in Fig. 11(a).For the in-band IIP3 measurements, two tones f 1 and f 2 are at f − 2.5 MHz and f + 2.5 MHz, respectively.Note that f represents the offset from the LO frequency (f LO ) and f LO = 2.5 GHz in the

B. VM Measurement
Relative amplitudes and phase shifts of all the possible VM settings are measured with the UXA N9040B I/Q analyzer option using the setup shown in Fig. 8.The UXA splits the measured signal into I and Q vectors.This measurement uses a single RF tone with a 50-MHz offset from the LO at 2.5 GHz.By taking the first measurement as a reference, all the control words of the VM were swept and the resulting I/Q vector constellations are shown in Fig. 11(b).The rms phase error of the VM is 1.48 • , the rms amplitude error is 0.33 dB, and the largest circle in the constellation has rms 8 • of resolution.The outer circle in Fig. 11(b) will be used both for rejection and calibration methods.

C. Modulated Measurements, Image Rejection, and EVM
Signal quality and I/Q image rejection were characterized with EVM measurements.The 5G NR 50, 100, and 200 MHz [2 × 100 MHz per component carrier (CC)] signals with cyclic prefix OFDM (CP-OFDM) FR2 were used in the measurements.An M8190A arbitrary waveform generator (ARB) by Keysight is used as a source to generate the wideband signal, and an E8257B vector signal generator (PSG) upconverts the signal to 2.55-GHz IF frequency (Fig. 8).The DSOS404A oscilloscope combines the IF-RX chip I/Q output at BB. Leakage signal suppression was measured only feeding input V Ar , while V Ai was terminated.A 16 QAM 50-MHz modulated  signal is generated for the test, and after deembedding the fixture losses, the power level at the IF-RX chip input is −25 dBm.Fig. 12 shows 33-dB suppression of the leakage signal.
Conductive EVM measurements were performed for a single-input case using 16 QAM and 64 QAM modulations.For 50-and 100-MHz signals, the RF is set to a frequency offset of 60 MHz from the LO, and for 2 × 100 MHz CC signal, the LO is in the middle of the CCs.The measured EVM curves as a function of the input power are shown in Fig. 13, illustrating that the IF-RX chip has more than 50 dB of dynamic range.The best achieved EVM is 1.4% for 50 MHz, 2%-2.2% for 100 MHz, and 3%-4% for 2 × 100 MHz wide signals.

VI. OTA MEASUREMENTS
In this work, the IBIC scheme is introduced to provide interfering signal cancellation in the analog domain.This will cause a decent reduction in the required ADC dynamic range and reduces the need of cancellation in the digital domain.The OTA measurements are performed to show the effectiveness of the spatial filtering with the IBIC scheme using EVM and beam pattern as in [21] and [33].IBIC was initially demonstrated using a continuous-wave (CW) signal in [21].In this section, the IBIC measurements were performed for wideband signals using the setup, as shown in Fig. 14.Wideband cancellation includes the frequency dependence not visible in many earlier state-of-the-art papers, in which the signal bandwidth is 50 MHz or below [12], [13], [15], [16], [17], [18], [19].The measured IBIC results are compared with the OTA theoretical IBIC model for wideband signal described in Section IV-B.The focus is on the nonidealities of the system and delays in the radio channel that limit the cancellation BW Canc .

A. mmWave OTA Setup and Link Parameters
The modulated signal OTA measurements were performed using a setup shown in Fig. 14.The measurements used two M8190A ARBs, I/Q upconverters, conventional PAs (CA2630-141), and horn antennas (A-info LB-28-15) on the transmitter side.A DSOS404A oscilloscope was used on the receiver side to measure the I/Q signal at BB.The 28-GHz front end of [34] has two 16-× 4-element subarrays (A1 and A2) with 16 electrically controllable phase shifters, one for each 2 × 2 hard-wired antenna subarray.The common path of each subarray has a 31-dB range 5-bit digitally controlled attenuator.Physically, A1 and A2 are horizontally aligned on top of each other, and they provide an IF output from 3 to 4 GHz.The outputs are combined at BB using the IF-RX chip shown in Fig. 14.A single input V Ar of the IF-RX chip is used, and the other input is terminated to 50 because the mmWave subarray provides a complex weighted IF output.
To evaluate the system gain and NF parameters, a 2-m OTA link with subarray A1 and IF-RX1 on and subarray A2 and IF-RX2 off is analyzed.At 28 GHz, the signal has 61 dB of loss for a 1-m distance in the OTA channel.For effective isotropic radiated power (EIRP) of +17 dBm at the transmitter antenna, the input at the receiver antenna array is −50 dBm.After combining at mmWave before the mmWave attenuator, the front-end subarray has a typical gain of 35 dB, including a 12-dB array gain.The downconversion from 3.5-GHz IF to BB has a 2-dB loss, including cable, PCB, and BB combiner losses and gain of the IF-RX.The attenuator in the signal path is set from 17 to 20 dB to adjust for optimum performance of the 28-GHz front end, resulting in 14-17-dB gain from the antenna array to BB output I port having NF of 9 dB.In the case of modulated measurements, the system has an additional 3-dB I/Q combining gain in BB.
In Fig. 14, the signal (blue) is at 0 • direction and the interference (red) is at −10 • .The measurements were carried out with 5G NR 16 QAM and 64 QAM OFDM signals, and minimal EVM requirements for demodulating the signals are 12.5% and 8% [35].The signal and interference are transmitted at 27.9 GHz.The EIRP of the transmitted signal is +17 dBm, and the interference EIRP is adjusted ≈10 dB stronger than the signal.On the receiver side, A1 is steered to 0 • and A2 to −10 • to receive both of them.The A1 and A2 mmWave subarrays utilize an in-house developed beamforming calibration scheme [36], which equalizes the main beam gain at different angles and reduces the SLLs below 13 dB for specific steering angles.The 3.5-GHz IF output is downconverted to BB with a center frequency of 60 MHz for 50-and 100-MHz wide single-carrier (SC) OFDM signals and zero offset for multicarrier signals.The BB I/Q outputs were combined using a DSOS404A oscilloscope.

B. OTA Modulated Measurements
In OTA modulated measurements, we will demonstrate: 1) IBIC after calibration of IBIC VM-coefficients; 2) OTA IBIC model validity and how we obtained branch mismatch and delay parameters to fit the model to measurements; and 3) wideband IBIC in a 5G NR carrier aggregation system.
Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.  1) Calibration of IBIC Coefficients: A 16 QAM 50-MHz modulated signal is used to demonstrate the calibration of complex cancellation coefficients.We first calibrated the amplitude matching between the main and cancellation paths.It requires information of the spatial filtering of the interferer by mmWave subarray.In this measurement, the interferer is ≈10 dB stronger than the signal, and its first sidelobe is in the main beam of the signal, as shown in Fig. 15(a).The interference and signal levels are measured at BB of the A1 receiver chain to verify spatial filtering by activating one TX at a time.During these measurements, the VM of the IF-RX1 and IF-RX2 chips is set at the outer circle and provides no phase rotation.The mmWave subarray spatially filters the 10-dB stronger interferer, and the received interference SLL is 7 dB below the signal in the direction of 0 • [see the red curve in Fig. 15(a)].After spatial filtering, the signal and interference spectrums can be seen in Fig. 15(b), showing the 7-dB SINR that corresponds to 34% EVM.After this measurement, the amplitudes are matched between A1 and A2 by using the attenuators in the mmWave subarray modules.The attenuation in the cancellation path depends on the difference between the main beam and the SLL of the interferer and is adjusted Fig. 16.
Calibrating the phase of the cancellation coefficients using 16 QAM-(a) phase1: interference power level after rejection is used to find the best cancellation coefficients.Lowest values indicate the lowest interference level-(b) phase2: fine-tuning of the control word is done by minimizing the EVM when the interferer is present.
accordingly.The measured beam pattern of the interferer in Fig. 15(a) (red curve) suggests that for amplitude matching, the difference between ATT1 and ATT2 is 18 dB.
In modulated measurements, the phase cancellation coefficient calibration method described in Section IV-C is performed in two phases and validated.Both TXs and RXs are turned on for calibration of the cancellation coefficients.A1 receives the signal with interference.In the measurements, attenuators in A1 and A2 are used to match the amplitude in the cancellation path, allowing cancellation to be done with the outer circle of the VM (better resolution).In the first phase, the channel power is measured at 0 • , 90 • , 180 • , and 270 • phases by changing the VM quadrant (IF-RX2) phase controls, as shown in Fig. 16(a) (shown in blue markers).The fine-grained VM phase controls are then searched toward the lowest power level.The lowest power value in Fig. 16(a) indicates the lowest interference level, and the maximum value corresponds to the maximum interference level.Based on the signal plus interferer measurements in BB, the required cancellation phase can be approximated for minimizing the interference.
The estimated cancellation phase does not always correspond to the minimum EVM because the rejected interference also has a fraction of the signal.Therefore, the EVM is fine-tuned by recalibration with N -points close to the cancellation phase in the second phase.Based on the channel power measurements as shown in Fig. 16(a) (magenta points), the calculated cancellation phase for optimized EVM is −72 • according to ( 6) and (7), which corresponds to VM phase control word 35.Measurement results in Fig. 16(b) (magnified from 31 to 38) confirm that the channel power-based initial estimate is, in this case, also equal to the control word of the optimized EVM.The digital demodulation constellation after IBIC is shown in Fig. 15(d).The amplitude matching, phase finding, and fine-tuning of the EVM measurements were automatized and take 11 measurements to complete one iteration.Most of the time best EVM is achieved in one iteration.
2) All Branch Mismatch Errors: The nonidealities in the measurements are estimated using the OTA IBIC model.The OTA IBIC model uses the same branch mismatch parameter values without delays given in Section IV-B, tuning A, OTA delay between branches (τ 2 − τ 1 ), and cancellation frequency to match the model with measurements.The additional multipath reflection signal is added in the model to estimate the delay in the path of reflection (τ 3 ).The OTA IBIC model is matched with measurements in three steps: matching null by tuning A and the cancellation frequency without OTA delays in the first step; then include delay difference between branches and multipath reflection delay and magnitude, and the null matching provides the estimated I/Q imbalance; and the second and third steps model the rejection spectrum and BW Canc by estimating OTA channel delays.
Fig. 15(b) shows that the signal and the overlaying interferer without IBIC have both a flat frequency response over 50 MHz of signal bandwidth.In the OTA IBIC model, the interferer level is set to the measured level of interference before the rejection.Fig. 15(c) shows the measured signal and interference spectrum after rejection with the simulated rejection (light blue curve).The simulated response in Fig. 15(c) is a combination of all branch mismatch errors, including OTA channel delay between branches and multipath reflection components.It is achieved in three steps as described.The null matching is achieved by tuning the center frequency and A difference to 4% without delay between branches.This results in 0.37-0.4dB amplitude and up to 2 • phase error.After combining in BB, this I/Q imbalance error corresponds to ≈0.9-dB amplitude and up to 6 • phase error.This parameterization shows good agreement with the measured null depth of 25 dB shown in Fig. 15(b).
OTA channel delays are included in the simulated model to match with the measured spectrum after rejection.OTA channel delay difference (τ 2 − τ 1 ) depends on the spacing between the arrays and the angle of the interferer.The τ 2 − τ 1 value can be calculated using simple trigonometry for an LOS channel.In the testing configuration, the spacing between arrays is 22λ at the carrier frequency, and the interferer is at an angle of about ≈11 • , resulting in a ≈153-ps delay difference.The multipath reflection component is added to the OTA IBIC model, and the delay τ 3 , the multipath reflection amplitude, and the cancellation frequency are tuned in the model to match the measured rejected interferer spectrum in Fig. 15(c).The resulting magnitude of the multipath component is 30 dB below the interferer, and τ 3 is ≈20 ns.The combined simulated effect in Fig. 15(c) (light blue) shows that the OTA IBIC model can be used to estimate the nonidealities in the system with good accuracy.Fig. 15(c) shows the achieved cancellation BW Canc above 20-dB rejection is 30 MHz and more than 50 MHz for 15-dB rejection.The measured demodulated signal constellation after rejection is shown in Fig. 15(c).After rejection, 6.8% rms EVM is achieved.The interference rejection in terms of channel power can be quantified from the measured beam pattern of the interferer after rejection [see Fig. 15(a), magenta line], which shows 19-dB rejection.At the rejection null, the SINR is mainly limited by the noise floor of the signal in this measurement, which results in SINR of 23-25 dB and rms EVM of 6.84%, which can be verified from the subcarrier rms EVM spectrum, as shown in Fig. 15(i).The reliability of the calibration method and measurements is double-checked with 64 QAM signal, and the results are Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.shown in Fig. 15(e)-(h) and (j) showing similar rejection performance.
3) Reflections at mmWave in the OTA Channel: Another set of interference rejection measurements was performed for a 100-MHz wide signal to evaluate the impacts of multipath components in the radio channel.Those are causing differences between the cancellation path and the interference path.These are frequency dependent and seen as ripples in the response, as shown in Fig. 17(b).This is verified with simulations by changing the cancellation frequency in the OTA IBIC model and keeping all the other parameters the same as described in previous OTA 50-MHz signal measurements.The simulated response can be seen in Fig. 17     4) Wideband Rejection in Carrier Aggregated Systems: As relative bandwidth increases, frequency dependence becomes even more prominent.In broadband systems, a single-tap analog canceller as shown above has fundamentally limited cancellation performance over the spectrum.One way to improve the rejection bandwidth is to divide the processing into multiple separate channels to equalize the frequency response and match the OTA channel delays in subbands.For example, in 5G NR, carrier aggregation schemes are used to divide the overall band into multiple CCs.To illustrate this mode of operation, a 4 × 100 MHz signal is used in the measurements to demonstrate multitap performance over several CCs in OTA cancellation.Fig. 18 shows the frequency spectrum of the signal and the interferer.From CC0 to CC3, the signal amplitude varies by 1.5 dB, and the interferer amplitude varies by 2.7 dB.It is because the IF-RX chip has a nonflat response around 3.5-GHz input and some additional in-band ripple in the IF BPF of the mmWave front end [34].Under these constraints, single CC cancellation can be accomplished by matching amplitudes with attenuators in the mmWave front end and calibrated each CC separately in the cancellation process.It is shown for each CC separately in Fig. 19.For more than 15-dB rejection needed for successful digital demodulation of each CC, the amplitudes for a particular CC are adjusted at a time.In that case, the amplitude matching error is between 0.8 to 1.1 dB, and the multipath reflections delay is between 17 and 20 ns.Fig. 19(a)-(d) shows the measured rejection spectrum and simulated rejection response of each CC by matching the amplitude of single CC at a time.In the simulations, all the parameters were the same except the delay of the multipath reflections for each CC and matched the response by tuning cancellation frequency and delay of the multipath reflection to match each other.The achieved rms EVM for each CC after rejection is from 7.3% to 9.3% [Fig.19(e)-(h)], and the rejection in terms of channel power is from 16 to 17 dB.Better performance would require four parallel cancellers one for each 100-MHz CC in this case.

C. Cancellation Performance Comparison
Comparing wideband cancellers for realistic OTA scenarios is never a straightforward task.As shown in this article, cancellation always has limited bandwidth, not only depending on the relative frequency or the cancellation technique but also on the radio channel and how the rejection has been tested.Many of the cancellation schemes referred to in this section are working fine for the CW tone or low bandwidth cases, but they have not been tested OTA with wideband signals.Table I summarizes the performance of the IF-RX  chip and rejection scheme and benchmarks it with state-of-theart cancellation techniques.The interference rejection results include the spatial filtering of the external mmWave module in front of the IF-RX chips and IBIC.A comparison is made with mmWave wideband and multiband [12], [13], RF front end only [15], and sub-6-GHz [17], [18] receivers.The IF-RX Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.achieved 1.8-GHz IF bandwidth, which is limited by external balun, and has the widest BB bandwidth of 400 MHz, making it suitable for wideband combining or interference rejection in mmWave 5G NR systems.Huang and Wang [13] covered most of the mmWave 5G NR band and has 2.5-GHz IF bandwidth but has the largest chip area per cancellation stage because wideband cancellation is achieved by creating delays with passive components at IF and mmWave.IF-RX chip has higher power consumption compared to others (single element) due to differential implementation and Cartesian combining function, which requires four differential stages in one slice.
Please note that we use realistic OTA measurement scenarios and do not use a frequency offset.This work achieves 34-37 dB interference rejection when IBIC operating as part of the mmWave system at the worst case sidelobe scenario.In spatial filtering with a uniform array, 18 dB comes from the array, and the rest comes from the IBIC.It shows the effectiveness of this architecture for wideband signals.On the other hand, Mondal and Paramesh [12], Huang and Wang [13], and Jain et al. [15] used conductive measurements for interference rejection, whereas Zhang et al. [16] used OTA and Golabighezelahmad et al. [18] used both OTA and conductive measurements.The measurements in [13], [17], and [22] used frequency offset except [18].However, conductive measurements show the wideband rejection but do not emulate the nonidealities of OTA channel, Golabighezelahmad et al. [18] showed 27-dB rejection over 100-MHz 16 QAM SC wideband signal cancellation in conductive measurements but only 15-dB rejection for 1-MHz 256 QAM SC signal in OTA measurements.Zhang and Krishnaswamy [17] used an 11-dB stronger CW interferer and showed 24-dB rejection.The conductive measurements of wideband rejection in [13] used passive delay lines at IF to match the delay differences between antenna elements.However, the rejection requires large angular separation, a minimum 3-dB power offset, and a 50% frequency offset of the signal bandwidth.
The reported data rates of OFDM signals were estimated using 3GPP TS 38.306 standard signals with all the slots used for the downlink and, for other references, reported the theoretical limit.Standard signal configurations in 3GPP are not corresponding to the theoretical maximum for certain data rates.Therefore, the given numbers are indicative only and not directly comparable to ones without protocol overhead.

VII. CONCLUSION
This article presented a subarray-based IF downconversion receiver architecture supporting hybrid beamforming for large-scale arrays with wideband IBIC.It utilizes both spatial rejection of mmWave subarrays and IBIC by rerotating the interfering beam to cancel the interference from another independent subarray.Subarrays also filter multipath reflections that cause frequency-dependent interference.However, even small reflections and RF nonidealities have a significant impact on cancellation properties of wideband signals.A theoretical OTA IBIC model with different nonidealities is analyzed and matched to OTA measurement results of the prototype receiver consisting of two mmWave subarrays and IF-RX chips providing IBIC.A two-phase calibration method to optimize the cancellation performance is utilized to achieve accurate and phase-optimized cancellation coefficients.
It was demonstrated in the OTA measurements that a single IF-RX with an attenuator to adjust amplitudes in the cancellation path achieves 16-19-dB rejection over 100-MHz bandwidth signals.Furthermore, for wideband carrier-aggregated waveforms, it is shown that cancellation can be trained for each CC individually, and 16-dB rejection is achieved over 400-MHz multicarrier 5G NR waveform.In contrast to many of the reported cancellers in the literature, our results are achieved for an interferer that is fully overlapping with the received data stream and realistic OTA signal conditions.

Fig. 1 .
Fig. 1.(a) Subarray per user approach with the cross-user IBI.(b) Proposed dual-stage mmWave phased array architecture for a large subarray receiver with parallel IF-RXs working as VM used in the main path (M) and cancellation/combining path (C) to realize IBIC.(c) Same IF-RX chip used to support an mmWave Cartesian combining architecture.

Fig. 2 .
Fig. 2. IF-RX chip architecture.The current combining at the output is done by a BB-TIA.
The four-phase 25% duty cycle current mixer converts the IF transconductor current into zero-IF, performs the Cartesian combining for V Ar and V Ai signals, and provides the differential I/Q signals.The phase selection block at BB rotates the I/Q signals by 0 • , 90 • , 180 • , or 270 • depending on the phase constellation point.

Fig. 3 .
Fig. 3. Implementation details of a slice of the IF-RX shown in Fig. 2: (a) Cartesian combining with gm-block and passive mixer, (b) gm/transconductor stage, and (c) phase selection block (SW).

Fig. 6 .
Fig. 6.(a) Simulated BB responses of branches 1 and 2 with mismatch parameters.(b) I Canc and Q Canc with and without delay difference (τ 1 − τ 2 ) between branches (c) combining I Canc + jQ Canc and (d) BW Canc for minimum 20 dB of cancellation with different error in A.
. The external passive balun (MTX2-73+) used at the RF and LO ports has 2-7-GHz bandwidth.Below 2.5 GHz, the external balun restricts the bandwidth of the wideband RF response in measurements.The RF frequency response of IF-RX chip is measured at 50-MHz offset across LO frequencies from 1 to 4 GHz.The IF-RX chip has a maximum conversion gain of 15 dB at 2.5 GHz shown in Fig. 9(a).Fig. 9(b) shows the measured BB response at a Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

Fig. 7 .
Fig. 7. (a) Chip photograph of IF-RX with power consumption of each block.(b) PCB photograph.

Fig. 12 .
Fig. 12. Measured IRR: input to V Ar and V Ai terminated.

Fig. 14 .
Fig. 14.(a) OTA modulated measurement setup and lab photograph of the setup in an anechoic chamber.(b) Placement of transmitters and A1 and A2 for calculating the radio channel delay difference (τ 2 − τ 1 ) between A1 and A2 receiver.

Fig. 15 .
Fig. 15.(a) Signal and interferer beam and interferer rejected at 0 • , (b) measured 16 QAM 50-MHz signal and interference at the output IF-RX1, and (c) signal after interference rejection.In (b) and (c), the measured spectrum before and after rejection is compared with OTA IBIC model (d) EVM with interference rejection (e)-(h) 64 QAM 50-MHz signal with interference rejection measurements.Spectral representation of rms EVM subcarriers of (i) 16 QAM and (j) 64 QAM signals.

Fig. 17 .
Fig. 17.(a) 16 QAM 100-MHz signal and interference at the output IF-RX1 and (b) interference rejection.In (a) and (b), the measured spectrum before and after rejection is compared with the OTA IBIC model.(c) Constellation with interference rejection.(d) Signal and interferer beam with and without interference rejection.(e) Rms EVM spectrum of each subcarrier.
(a)  and (b).The OTA IBIC model with delayed multipath component (τ 3 = 20 ns) confirms the ≈50 MHz periodicity in the rejected spectrum due to multipath reflections in Fig.17 (b) and shows ≈15 dB of minimum rejection over 100-MHz bandwidth.Fig.17 (c)  and (e) shows the corresponding demodulated signal constellation and EVM spectrum.Wideband rejection results in 7% rms EVM and achieved 17-dB rejection in terms of channel power when integrated over all subcarriers (see Fig.

TABLE I PERFORMANCE
SUMMARY OF INTERFERENCE CANCELING RECEIVERS AND COMPARISON