Extending 5G TDD Coverage With XDD: Cross Division Duplex

In this paper, an advanced duplex scheme called cross-division duplex (XDD) is proposed to enhance uplink (UL) coverage in time division duplex (TDD) carriers by utilizing self-interference cancellation (SIC) capability at a base station. With XDD, it is possible to combine TDD’s ability to efficiently handle asymmetric UL and downlink (DL) traffic with frequency division duplex’s coverage advantage. To do so, XDD simultaneously operates UL and DL on the same TDD carrier but on different frequency resources. Such operation leads to severe interference on the received UL signal at the base station which requires two levels of SIC implementation; antenna and digital SIC. More than 50 dB of interference is removed through the antenna SIC using electromagnetic barriers between the transmitting and receiving antennas. The remaining interference is removed by the digital SIC based on estimating the non-linear channel of the circuit at the receiver baseband. It is verified by simulation and analysis that with the proposed XDD, the UL coverage can be improved by up to 2.37 times that of TDD. To check the feasibility of XDD, a Proof-of-Concept was developed where it was observed that the benefits of XDD can indeed be realized using the proposed SIC techniques.


I. INTRODUCTION
Recently, the first release of the fifth generation (5G) standard, called new radio (NR), was completed in the 3rd Generation Partnership Project (3GPP) [1]. NR has been designed to support new use cases having a wide range of requirements such as enhanced mobile broadband, ultrareliable low-latency communications, and massive machinetype communications [2], [3]. Compared to 4G Long Term Evolution (LTE), NR supports operation on higher carrier frequencies up to tens of GHz, a larger bandwidth up to 400 MHz, and a larger number of mandatory receiver antennas up to 4 at the user equipment (UE) to meet higher requirements such as peak data rate up to 20 Gbps and 1 ms latency [4].
The associate editor coordinating the review of this manuscript and approving it for publication was Min Jia .
One critical issue identified during the initial deployment of NR systems was the limited coverage [5]. Due to the higher frequency bands used for 5G in order to support wider bandwidths, a larger path loss is inevitable [6]. Another factor that impacts the NR coverage is that majority of the new 5G spectrum allocations around the globe are time-division duplex (TDD) carriers located at around 3.3 −3.8 GHz, 28 GHz, or 39 GHz which are much higher than that of 4G [7].
TDD generally has a number of advantages over frequency division duplex (FDD). The first advantage is that time resources can be flexibly assigned to uplink (UL) and downlink (DL) considering the asymmetric ratio of traffic in both directions. DL is typically assigned the majority of the time resources to handle DL heavy mobile traffic [8], [9]. Another advantage is that channel state information (CSI) can be easily acquired via channel reciprocity 51380 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ VOLUME 9, 2021 FIGURE 1. Two main aspects to coverage reduction in 5G. One is a decrease in transmission time due to the use of TDD and the other is a decrease in the propagation distance due to an increase in the center frequency.
so the overhead associated with CSI reports can be significantly reduced especially when there is a large number of antennas [10], [11].
Although there are advantages of TDD over FDD, there are also disadvantages. The first disadvantage is the limited coverage due to the relatively small portion of time resources for UL transmission [12]. By assigning the majority of time resources to the DL, only a small portion of time resources can be allocated to the UL resulting in a smaller coverage (see Fig. 1). Note that FDD does not have this issue since UL has access to all time resources. 1 Another disadvantage is latency. In TDD, the timing gap between DL reception and UL transmission containing the hybrid automatic repeat request (HARQ) feedback of the corresponding DL is typically much larger than that in FDD [14], [15]. As a consequence, the HARQ round trip time in TDD is relatively longer than that of FDD, especially when the DL traffic load is high and causes a throughput loss.
There has been quite a bit of research in the wireless industry and academia [16]- [20] to address the shortcomings of TDD. One outcome of such research is the so-called supplementary UL (SUL) which utilizes a UL carrier in a lower frequency band in addition to a 5G carrier on a higher frequency band [16], [17]. The disadvantage of using SUL is that it incurs an additional cost in the form of an additional lower frequency carrier and implementation complexity increases at both the base station and the terminal side. Furthermore, SUL operation does not solve the fundamental problem of TDD but tries to avoid it by the use of an additional carrier. An alternative to SUL is to increase the density of 5G base stations [18], [19], [20] which would require more cell sites. From mobile network operators' (MNO) point of view, it is not attractive due to the additional economic burden. Furthermore, simple densification of 5G networks still does not address the issue of increased latency of TDD.
Another approach that could achieve coverage enhancement as well as improved latency is full-duplex (FD) [21]- [25]. By using FD at the base station, the UL and DL signals are received and transmitted on overlapping frequency resources [21]- [25]. Theoretically, spectral efficiency can be doubled and latency could be minimized. However, FD faces a number of issues in real life application. First issue is that due to using overlapping frequency resources, the received UL signal is subject to co-channel cross-link interference (CLI) [21], [23]. CLI cancellation methods include passive methods which rely on the antenna isolation between transmit and receive antennas, active methods which utilize RF or digital signal processing, and hybrid methods using a combination of these methods [43], [44], [45]. While handling co-channel CLI may require large complexity at the receiver side, it has been demonstrated by a number of researches that it is feasible. The more critical issue for FD is the handling of adjacent channel CLI. In many cellular band allocations, one operator's spectrum is located right next to the spectrum of another operator. For example, in one region of the world which now has nationwide 5G coverage, three MNOs are allocated 5G spectrum as follows.
• MNO#1: 3.42 GHz -3.5 GHz (3GPP band n78) • MNO#2: 3.5 GHz -3.6 GHz (3GPP band n78) • MNO#3: 3.6 GHz -3.7 GHz (3GPP band n78) Assume that MNO#2 is deploying FD in it networks. In such a case, it would have to handle not only the co-channel CLI within its own spectrum but also the adjacent CLI generated by the other two MNOs due to imperfect filtering on the DL transmission. In a conventional TDD system, such adjacent channel CLI would not be problematic since all three operators would be using the same UL-DL time resource configuration. In short, there would not be any adjacent channel CLI to begin with. Handling adjacent channel CLI is challenging due the difficulty of implementing active and passive cancellation. For active cancellation, the complexity at the receiver side would be extremely high since it requires the entire operation must be done without a priori knowledge of the DL transmitted signals of another MNO. Furthermore, for passive cancellation to work, the MNO who is generating the adjacent channel needs to implement interference suppression on it's own network for the benefit of the MNO who is deploying FD. In other words, for MNO#2 to deploy FD, MNO#1 and MNO#3 would need to share some of the financial burden.
This paper proposes an advanced duplex scheme, referred to as cross division duplex (XDD), and its base station implementation to overcome the coverage limitations imposed by TDD systems. XDD realizes simultaneous DL and UL operation within a TDD carrier by using different TDD configurations across different frequency regions. By doing so, XDD is able to adapt for different UEs individually with different TDD configurations. For UEs located at cell edge whose main concern is coverage, UL heavy TDD configuration is used to guarantee sufficient time resources on the UL to improve UL coverage. For UEs close to the cell center with good signal conditions, DL heavy TDD configuration is used to guarantee high DL throughput. These heterogeneous TDD configurations operate simultaneously at a base station within a single TDD carrier.
The performance of XDD is verified using numerical evaluation as well as actual hardware and software based Proof-of-Concept (PoC). It was observed that an XDD system utilizing UL time resource that is 5 times that of a conventional TDD system could extend the UL coverage area by a factor of 2. In the PoC, antenna SIC is used to minimize the leakage power from TX antenna to RX antenna. In addition, digital SIC is used to handle any residual interference after antenna SIC. It was observed that the DL out-band signal flowing into the UL receiver path can be effectively suppressed below the noise floor level to guarantee the UL receiver performance. Also, by combining digital predistortion (DPD) [37] at transmitter (TX) path and digital SIC at receiver (RX) path, the out-band interference from the DL signal to the UL can be effectively mitigated such that the guardband between the UL and DL signal can minimized.
The key contribution of this paper are the following: • A new duplex method, XDD, to address the coverage limitation of 5G TDD carriers. With XDD, the base station can improve the uplink performance at cell boundary regions and while being able to simultaneously handle DL heavy asymmetric traffic.
• A passive SIC method using multiple choke walls that is precisely designed to isolate co-channel CLI over a wideband. Unlike active SIC method where the complexity becomes prohibitive for large number of antennas, the proposed passive method is well suited even for massive MIMO [52].
• State of the art digital SIC to handle the remaining co-channel CLI. The digital SIC is designed to operate in conjuction with digital pre-distortion (DPD) and antenna isolation. The digital SIC provides a high degree of flexibility for XDD in that the UL and DL subbands can be flexibly located over the bandwidth depending on the need of the use cases.
The rest of this paper is organized as follows. The overall concept of XDD is discussed in Section II. Design principles of XDD system architecture are described in Section III. Simulation results are provided in Section IV. The feasibility and effectiveness of XDD are verified based on PoC results in Section V. Finally, conclusions are drawn in Section VI.

II. XDD CONCEPT A. REVIEW OF DUPLEX SCHEMES
Before discussions on XDD, it should be worthwhile to briefly summarize the duplexing schemes that have been used in conventional cellular systems. In early cellular systems such as 3G, FDD was dominantly used due to the narrowband nature of wireless signals [26], [27]. Although TDD was introduced to supplement FDD in 3G, it was not popular due to limitations in coverage [28]. However, with wireless traffic expanding exponentially, scale, especially due to DL heavy traffics such as wireless internet and video, the benefits of TDD became evident [13]. In fact, for many of the 4G systems utilizing FDD, the UL is often underutilized while the DL is under heavy use. In 5G, most of the new carriers are TDD carriers and therefore to address such issues [12]. Of course, the fundamental problem of coverage limitation of TDD is still not addressed.
There have been attempts to use TDD and FDD at the same time to combine the benefits of both duplexing schemes in proposals such as hybrid division duplex [29]- [31]. Although the approaches in [29]- [31] are easy in concept, realizing an actual cellular system based on such a concept is no easy task. There are a number of challenges that need to be addressed. First, within a base station, the transmitted signal causes self interfering co-channel CLI on the received signal. Second, between base stations of the same MNO, there is CLI due to the difference in duplex direction at a given time. Third, between base stations of different MNOs, there is adjacent channel CLI due to leakage signal from the neighboring channel. The purpose of this paper is to propose a new duplexing scheme using the current NR specifications that combines the benefits of TDD and FDD but designed so that the above challenges can be addressed effectively.

B. XDD: CROSS DIVISION DUPLEX
The concept of XDD is depicted in Fig. 2. XDD is a duplexing method in which duplexing can be implemented in either time domain, frequency domain, or both domains within a single TDD carrier depending on the needs of the deployment scenario. A base station can schedule non-overlapping resources (in frequency-domain) to terminals so that DL and UL transmissions occur even in the same time instance. In other words, a cell-edge terminal can be assigned to transmit continuously on UL resources (user #1 in Fig.2) while DL transmissions are being made at the base station side to serve other users (user #2 in Fig.2) at the same time. It is worth noting that the mobile station does not requires any additional implementation for XDD. As a result, the accumulated UL energy at the base station can be larger for XDD which would increase the coverage beyond what is possible for TDD. Note that since a typical cell edge terminal requires only a narrow frequency resource, the remaining portion of the carrier can be used for DL transmissions. As mentioned beforehand, such an operation requires advanced SIC capability at the base station to handle the self-interference. In addition, according to the presence of UL transmission, the base station can effectively receive the UL signal by abortively applying self-interference cancellation. One characteristic of XDD worth noting is that it requires no change on the terminal side. Conventional TDD terminals can be supported by a XDD base station without any modification.

C. REQUIREMENTS AND CHALLENGES FOR XDD IMPLEMENTATION
To implement XDD at the base station, the following requirements should be met. First, it is necessary to prevent analogto-digital converter (ADC) saturation [32], [33] due to the high power DL signal coupling to the UL path. When ADC saturation occurs, the received UL signal is severely degraded and cannot be restored even after digital SIC. Second, the residual interference must be reduced below the receiver's noise level before UL demodulation and decoding operation. In many cases, even a small level of interference left after SIC could make it impossible to use high order modulations such as 64QAM and 256QAM. Finally, the frequency separation between the UL and DL acting as the guard band should be minimized. If SIC is not properly implemented and there is a significant level of residual interference, the guardband between a DL signal and a UL signal would need to be increased causing a loss of resource efficiency (see Fig. 3).
There are a number of challenges in achieving the requirements described above, some of which are as follows: 1) Modeling of the non-linear self-interference channel: In order to effectively operate the SIC on the selfinterference which has time varying non-linear characteristics, active components (e.g., high power amplifier, cascaded power drivers) and passive circuits (e.g., filter) must be modeled accurately [34], [35]. 2) Interference oversampling and synchronization: In order to remove the interference signal, the relationship between the interference signal and the original signal must be closely observed. To this end, the receiver needs to perform oversampling at a rate of 5-7 so as to obtain perfect synchronization, which is very difficult for a system with large bandwidth. 3) Internal power coupling and crosstalk: No matter how much interference signal is reduced at the receive antenna, if the crosstalk between the components in the internal circuitry cannot be suppressed sufficiently, it deteriorates the performance of interference cancellation. 4) SIC method in the RF domain: In order to prevent ADC saturation at the receiver chain, it is necessary to suppress the signal as much as possible at the RF stage. The active cancellation method studied so far is capable of an acceptable level of cancellation, but the complexity is too high to be used for a bandwidth over 100 MHz and more than 32 multi-antennas (typical 5G base station). Therefore, a low-complexity RF SIC method that can be applied regardless of the number of antennas and bandwidth is required for commercial viability. 5) SIC method in the digital domain: Various digital SIC methods are studied in the literature. However, many state-of-the art solutions are impractical in implementations due to fixed-point processing and the high algorithmic complexity In order to actually use XDD, it is necessary to implement an algorithm that can be implemented in an FPGA to reduce residual interference. Based on these requirements and challenges, we propose a system architecture that maximizes the benefits of XDD in the following section. Figure 4 illustrates the SIC system model for typical TDD base station architecture. Let x(n) be the transmit baseband signal for time sample n. The non-linear PA output signal x nPA (n) can be expressed using parallel Hammerstein Model [36] as follows:

III. DESIGN OF XDD SYSTEM ARCHITECTURE A. SYSTEM MODEL
where P is the maximum non-linearity order, f p,n is the PA impulse response of length N −1 for the nonlinear order p, and the operator • represents the convolution operator. Expressing the coupling response from TX to RX chain (e.g., PA to Lownoise amplifier (LNA)) as h n , the leakage signal at the RX chain x lkg (n) can be expressed as where h p,n is the unknown coupling response for nonlinearity order p and time sample n. The received RX signal y(n) can be expressed as where x ul (n) is the desired UL signal and z(n) is the additive Gaussian random noise. To remove leakage signal x lkg (n) from the received signal y(n), the base station firstly needs to reduce the leakage power level to prevent ADC saturation and then, needs to emulate leakage signal from the transmitted signal. That is, whereĥ p,n,τ is the estimated leakage channel coefficient for time delay τ . Two dominant components which contribute to the leakage channel are surface wave leakage and near field coupling. The receiver can handle these two components together with a combined set of leakage channel coefficients. In the following subsection, three SIC steps are described for the XDD base station. These are adjacent channel leakage ratio (ACLR) reduction with DPD, surface wave and near field coupling cancellation with choke wells, and the digital SIC for handling residual interference at the base station.

B. SELF-INTERFERENCE CANCELLATION
Effective SIC is critical for XDD operation. We accomplish this with three different technologies that work together which we will detail in this section. Firstly, we linearize the transmit chain, reducing the ACP which falls onto the downlink band. Secondly, we improve the isolation in the antennas so that less DL interference will flow into the UL antennas. Lastly, we use Digital SIC to remove any remaining interference in the digital baseband.

1) STEP 1: ACLR REDUCTION USING DPD
The TX chain includes the primary PA as well as many predriver stages and filters. Each of these are necessary in a commercial product and contribute to the nonlinearities and frequency shaping experienced in the leakage. This cascade of nonlinearities in the TX Chain can be effectively modeled as a single nonlinear system shown in Fig 5. An increased linearity can be achieved by applying DPD on the transmit signal after which (1) can be rewritten as wherex(n) is the pre-distorted TX signal at the baseband. Typically, the generalized memory polynomial (GMP) model is widely used as the pre-inverse model for the nonlinearity of the systems, memory effects, and cross-memory effects [38]. The output of pre-distorted signalx(n) can be expressed as where a p,m , b p,m,l , and c p,m,l represent coefficients for polynomials for the pre-distortion. P a and M a are the index arrays for aligned signal and envelope. P b , M a , and L b are the index arrays for signal and lagging envelope. P c , M c , and L c are the index arrays for signal and leading envelope.
To maximize the linearity of the TX chain, the indirect learning architecture with least-squares (ILA-LS) is utilized [42]. Let β be a vector collecting all the GMP coefficients, e.g., a p,m , b p,m,l , and c p,m,l in (7). The key of ILA-LS is to learn the coefficients that minimize the LS errors between the output of the pre-distorted signalx and the post- whereX nPA is a matrix representation of the GMP forx nPA where columns correspond to basis functions of the polynomial. Thus the optimumβ can be expressed aŝ and we haveβ To do this, the output of PA signal,x nPA (n), is captured before the antenna port and fed into a GMP estimator to compare pre and post-distorted signals. Noting that this process requires an additional RX chain that includes a coupler, attenuator, and ADC. Onceβ is obtained, this vector is applied to pre-distortion and iteratively generatesx nPA (n) until convergence is achieved. To end this, Tikhonov regularization is used to improve the numerical stability of the LS calculation [39]. This approach will place a small penalty on the magnitude of each element of β and also makes it suitable for FPGA implementation.
Using the DPD, ACLR after PA can be reduced by more than 10 dB without degrading the error vector magnitude (EVM) at the TX chain.

2) STEP 2: SURFACE WAVE LEAKAGE AND NEAR FIELD COUPLING CANCELLATION WITH MULTIPLE CHOKE WALLS
After enhancing the linearity of the TX chain, the next step is to cancel the surface wave leakage between TX and RX antennas. To do this, a novel antenna SIC is used, which relies on multiple choke walls forming a corrugated plane surface. The structure of the choke wall is illustrated in Fig. 6. Since the surface wave is composed of transverse electric (TE) and transverse magnetic (TM) waves, both types of waves can be canceled. By cascading the perfect electrical conductor (PEC) and the perfect magnetic conductor (PMC), both TE and TM waves can be suppressed [41]. There are three design parameters for choke wall: height of each wall, the distance between TX/RX antenna and the closest choke wall d 1 , and inter-wall distance d 2 .
There are two approaches to design the wall height, the first is to use a height between λ 4 < d < λ 2 for achieving high attenuation using walls, and the second is to have a different height between walls to cancel the wideband signal. One example of using two choke walls is to have the height of the first and second walls as (λ + δ 1 )/2 and (λ + δ 2 )/2, respectively. The values of δ 1 and δ 2 are obtained to achieve the best suppression across the target bandwidth. Next, the distance d 1 should be considered to avoid the cancellation of the reflected wave and radiated wave at the TX antenna. To prevent 180 degrees of phase shift in the reflected wave, d 1 should be approximately 1 4 of wavelength. Lastly, d 2 should be selected by considering the thickness of the wall (t). When t d 2 + t < λ 2 , there will be one non-evanescent mode presented within the corrugations. Hence, the thickness of the wall should be thin enough and the distances between two adjacent walls should be less than half of the wavelength. All the parameters are obtained from high-frequency electromagnetic solvers (HFSS) to target the design requirement of the XDD base station. Together with the surface wave canceling, the near-field coupling should be minimized together. In general, as the distance between the TX and RX antennas is increased, the coupling effect is decreased. In addition to coupling power reduction, the number of choke walls can be increased to reduce surface wave even more. However, considering the form factor of antenna port, the dimension of the antenna part will be more than double. From the measurements from XDD PoC, with approximately 4 times of wavelength separation between TX and RX antennas, there can be at least 5 walls. As a proper design of Walls, in total 55 dB isolation can be observed between TX and RX antenna. Isolation performance of choke wall and separation contributes 10 dB and 45 dB with the same polarization, respectively.

3) STEP 3: DIGITAL SIC FOR REDUCING RESIDUAL LEAKAGE POWER
Once DPD and antenna SIC are utilized, the next step is to estimate the leakage channel coefficientĥ p,n,τ and delay τ for canceling out the remaining leakage signal from the received signal. As depicted in Fig. 7, another GMP model is used to estimate the non-linearity of the overall leakage channel. Instead of calculating a pre-inverse model of the PA in the TX, the goal of leakage channel estimation is to model the nonlinear channel from the PA to RX chain. Given that the input and output signals of the PA are known, the system can be solved directly by LS in (10).
Note that, unlike FD, the received UL signal interferes with the adjacent leakage signal of the DL TX. This infers that when the ACL power level of the DL signal is suppressed below the level of noise floor by step 1 and 2, UL performance will not be degraded without applying the step 3 SIC. On the other hand, if the ACL power level is relatively higher, all interference power received at the RX path must be removed.  BW is not used. At the RX chain, remaining 20% is used for UL reception. The values of X and Y are design parameters, which are antenna SIC and ACLR performance, respectively. To compare the results between the SIC budget analysis and PoC, 91 dB pathloss is set as a reference. The TX and RX antennas are designed to have RF isolation of at least X dB by using dedicated antennas and isolation walls as described in Section IV. The TX chain has various nonlinear components, specifically the cascaded power drivers and the high-power PA that cause adjacent channel leakage (ACL) to out-band (unused band) where XDD UL reception will operate.

IV. XDD PERFORMANCE ANALYSIS A. SIC BUDGET FOR XDD OPERATION
Based on the design parameters of X and Y, the necessary region for digital SIC and the required digital SIC budget can be depicted in the upper and lower sub-figure in Fig. 8, respectively. The blue region in the upper sub-figure indicates the region where the remaining residual interference is lower than the noise figure level. In this case, digital SIC is not required and the UL signal can be decoded without interference. Otherwise, the total DL leakage power interferes with the UL signal and the receiver requires digital SIC in the in-band region. For example, if 55 dB isolation is achieved with antenna SIC with 50 dBc ACLR, the required SIC at the digital level would be 43 dB. Meanwhile, when ACLR is increased to 60 dBc, digital SIC is not necessary at all.

B. UL COVERAGE ENHANCEMENT WITH XDD
To verify the coverage extension performance with XDD, link level simulation has been performed. The carrier frequency and subcarrier spacing are assumed as 3.5 GHz and 30 kHz, respectively. To consider the coverage limited environments of cell-edge terminals, the number of allocated resource blocks (RBs) and the corresponding occupied channel bandwidth was assumed as 4 RB and 1.44 MHz. Other parameters for the link simulation are shown in Table 2. For TDD UL, only one packet is transmitted while 5 repetitions are used for XDD UL. For repetitive transmission of XDD, the IR (incremental redundancy) based HARQ is applied [53]. From the results shown in Fig. 9, more than 7 dB of SNR gain is achieved for 5 times longer UL transmission with XDD.
Based on the link evaluation results, the link budget of XDD and TDD can be summarized in Table 3. The residual self-interference assumption of less than 1 dB is based on the actual measurement taken from XDD PoC. At the UE side, the actual radiated power from the transmit antenna (c) is the sum of transmission power (a) and the effective UE TX gain (b). At the receiver (i.e., base station) side, the effective noise power (f) can be obtained as the sum of thermal noise power and the effective noise and interference power for the data channel. The thermal noise power can be calculated by the multiplication of thermal noise density (d) and the occupied channel bandwidth (e.g., 4 resource blocks in NR system). Residual interference after SIC is added in (e) for XDD. Then, two different required SINRs for TDD and XDD to satisfy the target BLER (g) for the given transmission bit rate are applied. As a consequence, the receiver sensitivity (h)  can be calculated by the sum of effective noise power (f) and the required SINR (g). Taking into account the effective base station antenna gain (i) and channel fading margin (j) properly, the hardware link budget margin (k) for each duplexing scheme can be calculated ((k) = (c)+(i)−(h)). Finally, available pathloss (n) can be calculated with shadow fading (l) and penetration margin (m), that is, (n) = (k) − (l) − (n). The analysis shows that the proposed XDD can extend the maximum UL radio distance by 54% compared with that of TDD. This radio distance improvement is equivalent to a coverage area that is 2.37 times as large as that of TDD.
A coverage area extension of 2.37 times means that the area covered by 7 conventional base stations using TDD can be operated with 3 or 4 base stations using XDD. In addition, if the coverage extension gain is transformed to the UL capacity increase, this result can be interpreted that the UL throughput can be improved by 4 times for mid-to-high SINR UEs.

V. XDD POC PROTOTYPING AND PERFORMANCE
A. XDD POC PROTOTYPE XDD PoC system is divided into two parts (see Fig. 10). The first part is the antenna part and the other part is the FPGA  part. In the case of the antenna part, there is one transmit antenna and one receive antenna, and each antenna has a 2 x 2 antenna element structure. Since the direct antennas can focus radiation energy in a certain direction toward the front, the amount of leakage signal at the UL antennas can be further reduced. In the case of the FPGA part, the SIC algorithm is implemented in Xilinx's ZCU111 [40]. Table 4 summarizes the XDD PoC design parameters. A 40 Gbps Ethernet link is connected between a host server and the FPGA platform. This link handles 32-bit data at 122.88 MSps from one DAC and two ADCs in real-time. The DAC receives data passing through ×32 up-sampling using interpolation filters. The output of ADC is passed through decimation filters that reduce the data rate by a factor of 32. The passband channel filters are used in the TX and RX chain to reduce the images, so as to have minimized the impact on the SIC performance. To estimate and increase accuracy of SIC over the wideband siganl, e.g., 100 MHz, 491.52 Msps is used.
To estimate channel between the base station and UE, the sounding reference signal (SRS) is transmitted.  To estimate leakage channel between DL to UL chain at the base station, DL reference signal is transmitted over DL only time slots before performing SIC. Detailed block diagram for XDD PoC and test setup in an anechoic chamber are depicted in Fig. 11.
To adapt parameters for DPD at the TX chain and SIC at the RX chain, at the PA output, a directional coupler is used to couple a small portion of energy through to the DPD/SIC feedback loop. A host server exchanges the IQ data samples using an Ethernet switch to place a pre-stored waveform that would play the output to the PA. The IQ data samples are computed based on the DPD polynomial and pre-distorted TX data as described in subsection III.B. The linearized PA output (after DPD) is now captured through the same path to verify performance improvement. For the UL reception of XDD operations, the RF switch is activated to provide access to the output of the LNA in the main path. The collected IQ data samples can be analyzed by a digital SIC algorithm to estimate the residual nonlinearity of the pre-distorted TX output. The residual from the digital SIC algorithm is analyzed at the host and performance metrics (e.g., EVM) computed.

B. XDD POC PERFORMANCE
The XDD PoC system is tested in an anechoic chamber with a UL signal from an emulated UE spaced 3 m away with power adjusted to simulate a cell-edge environment (see lower of Fig. 11). The DL TX power was set to 30 dBm. In the first step, both the UL and DL TXs are turned off to measure the RX noise floor. Then, the DL TX is turned on to observe the leakage power. Next, the process of DPD is executed to obtain the leakage power after DPD. Finally, the digital SIC method is conducted to obtain the residual DL cancellation power in the UL band.
The performance results are as follows in each step (see Table 5), for the performance of DPD applied in step 1, ACLR was between 40 −45 dB if DPD is not applied, which can be improved up to 50 dB with the proposed DPD implementation. The main reason that ACLR is made as low as possible is that digital SIC can be skipped depending on antenna SIC performance. Next, in step 2, the antenna SIC performance is observed. By introducing a total of 5 walls  with different heights, a distance of about twice the wavelength is required between the transmitting antenna and the receiving antenna. In the case of using the same polarization, it is possible to isolate signals of about 55 dB, and in the case of using different polarizations, 10 dB more isolation is achieved. As shown in Fig. 8, when the ACLR is 50 dB and the antenna SIC performance is 65 dB, the required digital SIC is 0 dB. In order words, using different polarizations, it can be seen that XDD can be implemented with very low complexity and without digital SIC. On the other hand, when the same polarization is used, the isolation performance is degraded by about 10 dB and a digital SIC is required for removing the whole DL leakage signal. As a result, a guard band is not required in XDD systems regardless of the digital SIC.
In order to check the performance of step 3, the worstcase using the same polarization is tested. In the case of TDD with only UL reception without DL, the received EVM of UL signal was observed to be about −24 dB. If the DL is simultaneously transmitted with UL reception without digital SIC, self-interference comes into the RX path and the EVM of UL signal is observed to be −14 dB, which degrades the performance by 10 dB. After applying the digital SIC proposed, the EVM performance is improved to a level of −23 dB, and interference is suppressed to less than 1 dB. The comparison of SIC performance is listed in Table 6 for antenna and digital SIC. For antenna SIC, various passivebased antenna cancellation or isolation techniques [46], [47], [48] shows 45 to 50 dB isolation without isolation walls while 65 dB isolation is achieved in XDD PoC. In the case of digital SIC, the SIC level of 50 dB is allowable budget for the digital SIC [49], [50], [51] with the limited ADC resolution. To handle SIC for both low and high modulation order, the budget for digital SIC should be minimized, which is 30 dB in XDD PoC. Figures 12 and 13 show power spectral density snapshots before applying digital SIC and after digital SIC, respectively. These results are measurements of the power spectral density in the frequency axis of the received UL signal observed in the RX chain. First of all, in the case of the situation without digital SIC (see Fig. 12), the interference power level of the  DL is higher than the UL signal level, and it can be seen that the interference to the UL actually interferes with the ACLR of DL signal, not in-band DL signal. Since ADC saturation does not occur due to attenuation by the antenna SIC, Only 10 dB degradation is observed in the EVM of the received UL signal. After applying the digital SIC (see lower sub- figure), the interfering DL signal has completely removed from the receiver. In addition, it can be seen that the SIC performance for the UL signal is significant even when there is no guard band between the UL signals of two DL subbands. The reduction in guard band is possible because the entire DL interference signal is eliminated.

VI. CONCLUSION
In this paper, XDD is proposed as a means to enhance the coverage of TDD carriers. XDD enables simultaneous operations of UL and DL on the same TDD carrier but on different frequency resources. Numerical evaluations, simulations, and PoC results show that XDD can extend the UL coverage area to be more than 2 times that of TDD. It was also observed that with proper implementation of SIC, it was possible to have negligible degradation on UL reception performance even when there is almost no guard band between UL and DL signals.
While only one transmit antenna and one receive antenna are considered in this PoC, it is worth noting that XDD PoC can be implemented with relatively low complexity even when a large number of antennas are used. In general, as the number of antennas increases, the pair between the transmit and receive antennas for SIC is proportional to the square of the number of antennas. As shown in this paper, if the isolation between antennas is sufficient with DPD, in XDD, it is possible to implement the multi-antenna XDD by applying digital SIC to only a few TX/RX pairs. In future work, we will extend the XDD platform to test with actual outdoor measurements. Since the importance of TDD is increasing in order to use a wider bandwidth than FDD, the proposed XDD can be a technology that can effectively reduce the cost of installing a cellular network while securing both the advantages of FDD and TDD at the same time. This advanced duplex technology will emerge as an important technology for beyond 5G and 6G in the future wireless communications.