Asymmetric Full-Duplex with Contiguous Downlink Carrier Aggregation

Transcription

1 Asymmetric Full-Duplex with Contiguous Downlink Carrier Aggregation Dani Korpi, Lauri Anttila, and Mikko Valkama Department of Electronics and Communications Engineering, Tampere University of Technology, Finland Abstract In this paper, a contiguous carrier aggregation scheme for the downlink transmissions in an inband full-duplex cellular network is analyzed. In particular, we consider a scenario where the base station transmits over a wider bandwidth than the mobiles, while both parties are still using the same center frequency. As a result, the mobiles must cancel their own selfinterference over a wider bandwidth, when compared to a situation where the uplink and downlink frequency bands are symmetric. Furthermore, due to the inherent RF impairments in the mobile devices, nonlinear modeling of the self-interference is required in the digital domain to fully cancel it over the whole reception bandwidth. The feasibility of the proposed scheme is demonstrated with real-life RF measurements, using two different bandwidths. In both of these cases, it is shown that the SI can be attenuated below the receiver noise floor over the whole reception bandwidth. I. INTRODUCTION Several recent works have shown that the concept of simultaneous transmission and reception on the same center frequency is practically feasible [1] [4]. The main challenge in implementing such inband full-duplex radios is canceling the own transmission from the overall received signal. This so-called selfinterference (SI) can be as much as db more powerful than the signal of interest, and hence advanced techniques are needed to solve this issue. As already mentioned, there are various demonstrator implementations, which have been able to tackle the problem of SI, rendering the inband full-duplex radio a feasible concept. For instance, in [1], a full-duplex relay prototype is presented and it is shown to cancel the SI almost perfectly. In fact, the whole principle of inband full-duplex communications is very well suitable for relay applications, since there the traffic is symmetric in terms of transmission and reception. This means that the inherent symmetry of an inband full-duplex transceiver is well utilized. Also the more generic implementations of inband full-duplex devices assume perfectly symmetric transmission and reception, at least in terms of the bandwidth [3] [5]. However, when considering a more practical deployment of a full-duplex device, assuming similar data rates for transmission and reception is typically not realistic. Especially, in a cellular network the uplink (UL) data rate requirements are typically much lower than the downlink () data rates [6]. Hence, in order to fully utilize inband full-duplex radios in a cellular network, a method for asymmetric data transfer is needed. In this work, this problem is addressed from the user equipment (UE) point of view, which means that the UL transmission data rate is assumed to be lower than the reception data rate. This can be achieved by having a wider bandwidth for the signal, and correspondingly a narrower bandwidth for the UL signal. For a high-quality full-duplex transceiver, this would be a trivial change since the SI cancellation procedure would only have to be performed over the transmit signal band, while the frequency bands outside that could be readily used for reception. However, since the focus of this work is on the UE side, the low quality of the RF components must be taken into consideration. In particular, the transmitter power amplifier (PA) will distort the transmit signal, resulting in a significant amount of spectral regrowth also outside the actual UL transmission band [5], [7]. This means that part of the SI is leaking to the adjacent frequency bands, calling for cancellation outside the intended transmission bandwidth. In this work, this type of a contiguous carrier aggregation scheme is laid out and analyzed, especially in terms of the required SI cancellation processing. Because of the aforementioned nonidealities in the transmitter, simple linear processing will not provide the required levels of cancellation, and thus nonlinear processing is needed to extend the reception bandwidth. The proposed scheme is then evaluated with real-life RF measurements, which incorporate also a state-of-the-art RF canceller. This means that the obtained results reflect the true overall performance of a mobile scale inband full-duplex device with asymmetric transmission and reception bandwidths. The rest of this paper is organized as follows. In Section II, the basic system model is presented, alongside with the full-duplex device architecture. After this, in Section III, the nonlinear modeling of the SI waveform is discussed, together with a detailed description of the nonlinear digital cancellation procedure. The RF measurement results are then shown in Section IV. Finally, the conclusions are drawn in Section V. II. SYSTEM MODEL AND DEVICE ARCHITECTURE This work focuses on analyzing a UE, which is assumed to engage in full-duplex communication with a base station (BS). Hence, the scenario is as illustrated in Fig. 1, where the situation is shown for several UEs. In the forthcoming analysis, however, only a single UE is assumed for ease of presentation. As already discussed, in a mobile cell the traffic between the UE and the BS is inherently asymmetric, since more data is typically transferred in the than in the UL. This is a fundamental challenge for an inband full-duplex system, which is usually assumed to transfer data, or at least deploy spectrum, in a symmetric way. In this article, a solution to this issue is proposed in the form of contiguous carrier aggregation. In particular, we /16/$31.00 c 2016 IEEE

2 FD-UE FD-BS FD-UE FD-UE Fig. 1. An illustration of a full-duplex BS communicating with full-duplex UEs in an asymmetric manner. UL 3rd order 5th order 7th order Fig. 2. A frequency domain illustration of the signal received by the UE under contiguous full-duplex carrier aggregation. The figure includes also the transmitter-induced nonlinear distortion up to the 7th order. assume a system where the primary UL and carriers are fully overlapping, but such that there are also additional carriers allocated for data transfer, which are directly adjacent to the primary or common UL/ carrier. Figure 2 illustrates this in the frequency domain, where the bandwidth is three times higher than the UL bandwidth. In an ideal situation, it would be sufficient for the UE to cancel SI only on the common UL/ carrier, since there are no ongoing transmissions in the -only carriers. This would result in an easier SI cancellation task, since the accurate regeneration of the SI signal is more challenging over wider bandwidths. However, as Fig. 2 illustrates, the ongoing transmission over the UL/ carrier produces SI also to the -only carriers due to the strong nonlinear distortion generated by the transmitter PA. If the SI cancellation is done only on the common UL/ carrier, the SINR of the -only carriers is heavily compromised due to this spectral regrowth. Hence, advanced digital SI cancellation techniques are needed to render the proposed contiguous carrier aggregation solution feasible. Simple linear digital cancellation procedures are of no use, since the linear component of the SI is not producing any interference at the -only carriers. Thus, in order to cancel the interference, a nonlinear model for the SI is needed, which is then used to regenerate the signal that is overlapping with the adjacent carriers. This can be done by using appropriate nonlinear behavioral models for the overall SI, and not limiting the cancellation processing and the underlying nonlinear basis functions to the purely inband part of the SI, which is the case f in the earlier works, e.g., in [5], [7]. Thus, the procedure is otherwise identical to the traditional inband SI cancellation, only the bandwidth is wider. Since this work involves SI cancellation in the mobile device, also the considered architecture for the full-duplex transceiver must be feasible for such a use-case. For this reason, we consider a full-duplex device, whose structure is as shown in Fig. 3. As can be observed, the full-duplex transceiver is sharing a single antenna between the transmitter and the receiver, which is typically a necessary feature for a mobile scale device. In addition, a three-tap wideband RF canceller is used to prevent the saturation of the receiver chain, in particular the low-noise amplifier. The small number of taps is made possible by utilizing the advanced canceller structure elaborated in more details in [8] and [9], since minimizing the complexity of the RF canceller is also a crucial aspect for a mobile scale full-duplex transceiver. After the RF canceller, the remaining SI is cancelled in the digital domain by the developed nonlinear digital canceller. Ideally, after this, the residual SI is well below the receiver noise floor at all carriers, and the full-duplex operation mode does not degrade the final signal-to-interference-plus-noise ratio (SINR) of the received signal of interest. III. SELF-INTERFERENCE SIGNAL MODEL AND NONLINEAR DIGITAL CANCELLATION In a mobile scale device, most of the nonlinear distortion in the transmit signal is typically produced by the PA [7], [10]. Thus, in order to cancel the nonlinear SI, a model for the PA must be incorporated into the digital SI signal model. A typical choice is to utilize the so-called parallel Hammerstein (PH) signal model for the SI observed in the digital domain [7]. Such a signal model has been shown to be accurate in modeling an actual low-cost PA under realistic operating conditions [1], [2]. Using the PH model, the observed SI in the digital domain can be expressed as follows: y(n) = P p=1 p odd M 2 m= M 1 h p(m)ψ p (x(n m)) + z(n), (1) where P is the nonlinearity order of the model, M 1 is the amount of pre-cursor memory, M 2 is the amount of post-cursor memory, ψ p ( ) is the pth-order nonlinear basis function, h p (m) contains the memory coefficients for the pth-order basis function, x(n) is the original digital transmit waveform, and z(n) represents the noise and possible model mismatch. Since the nonlinear canceller is modeling only the transmitter PA, which is producing the majority of the nonlinear distortion, it is sufficient to consider only the odd order basis functions. In order to use the above signal model for SI cancellation, the basis functions must first be generated such that the nonlinear SI also at the neighboring carriers is included, after which the corresponding memory coefficients must be estimated. Then, it is a straight-forward matter to perform the nonlinear digital cancellation. In the subsections below, a brief description of these issues is provided.

3 Wideband RF cancellation circuit Transmitter chain ττ 1 1 PA VGA IQ Mixer LPF DAC Transmit data BPF VECTOR MODULATOR VECTOR MODULATOR LO Receiver chain Nonlinear DSP Σ + LNA IQ Mixer LPF VGA ADC Σ + Digital cancellation To detector Fig. 3. A block diagram of the considered mobile scale inband full-duplex transceiver. A. Generating the Nonlinear Basis Functions In principle, the pth-order basis function is of the form ψ p (x(n)) = x(n) p 1 x(n) (2) when the nonlinear transformation is applied on the nth transmit data sample. However, this type of a presentation does not take explicitly into account the particular sampling frequency of the transmit signal x(n), which means that there might be some aliasing in the higher-order basis functions. In particular, the pth-order nonlinear transformation has a bandwidth p times as high as the original signal, which means that if the sampling frequency is not sufficiently high in the basis function generation stage, some of the nonlinear transformations will have frequency content higher than the Nyquist frequency. Especially, if the reception bandwidth is wide, this aliasing might even overlap with the signal of interest after the cancellation procedure. This type of a situation can be illustrated with the help of Fig. 2, where nonlinearities up to the 7th order are shown. In this case, if the Nyquist frequency in the basis function processing stage is significantly lower than the highest frequency components of the 7th-order distortion, some of the nonlinearities will alias on to the signal band, resulting in an inaccurate cancellation signal. As an obvious solution to this issue, the basis functions should be applied to an oversampled transmit signal, after which they can be low-pass filtered and decimated to the desired sampling frequency for SI cancellation. This ensures that no significant frequency content is aliasing on to the signal band of interest when generating the nonlinearly transformed signal with (2). This is an especially crucial aspect in the considered system, where the signal band of interest is much wider than the bandwidth of the actual transmit signal. In such a case, the transmit signal must be significantly oversampled to ensure that the nonlinear transformations within the final cancellation signal are not aliasing on to the reception bandwidth. Denoting the bandwidth of the transmit signal by B t and the bandwidth of interest by B r, the required sampling rate can be expressed as F s = P B t + B r. (3) 2 Using this sampling frequency for generating the basis functions ensures that no aliasing occurs within the reception bandwidth. From (3) it can be observed that the wider the bandwidth of the transmitted or received signal is, the more oversampling is required. Also increasing the nonlinearity order of the canceller results in a higher sampling rate requirement. For instance, setting P = 11 and having an IQ bandwidth of B t = B r = 20 MHz for both the transmit and receive signals, we get = 120 MHz, while having a received signal bandwidth of F s B r = 60 MHz results in F s = 140 MHz. However, since the highest order nonlinearities tend to be relatively weak outside the signal bandwidth, in practice a smaller sampling rate might also suffice. This is elaborated further in Section IV. B. LMS-Based Parameter Learning In order to apply the above signal model for SI cancellation, the coefficients in h p (m) must be estimated. To ensure a computationally relaxed learning procedure, the widely used least mean square (LMS) algorithm is utilized for this task. The procedure is similar to the one used in [2], where the LMS-based nonlinear canceller was successfully used for regular SI cancellation in a more elementary full-duplex device with symmetric UL and bandwidths. Now, the estimation and cancellation procedure itself is identical, meaning that the generated basis functions are first orthogonalized to ensure efficient learning, after which the LMS algorithm is applied to the orthogonalized basis functions. Denoting the orthogonalization matrix by S, which is obtained in a similar manner as shown in [2], the orthogonalized basis functions are defined as Ψ(n) = SΨ(n), (4) where Ψ(n) = [ψ 1(x(n)) ψ 3(x(n)) ψ P (x(n))] T, i.e., it contains all the basis function samples corresponding to the time index n. These orthogonalized basis functions are then used in the actual cancellation process, as follows: y c (n) = y(n) P p=1 p odd M 2 m= M 1 ĥ p,ort(m) ψ p (x(n m)) = y(n) ĥ H ortu(n), (5) where ψ p (x(n)) represents the orthogonalized pth-order nonlinear basis function from (4), and ĥp,ort(m) contains the corresponding estimates of the SI channel coefficients. The vectors

4 Circulator and antenna PXIe-5645R RF canceller TABLE I THE ESSENTIAL RF MEASUREMENT PARAMETERS. Parameter Transmit signal bandwidth (UL) Reception bandwidth () Center frequency PA gain Transmit power Value 20 MHz / 40 MHz 60 MHz / 120 MHz 2.46 GHz 23 db 6 8 dbm Number of taps in the RF canceller 3 Highest nonlinearity order (P ) 11 Number of pre-cursor taps (M 1 ) 15 Number of post-cursor taps (M 2 ) 15 Fig. 4. The RF measurement setup used for determining the integrated performance of the contiguous downlink carrier aggregation solution. are defined as ĥ ort = [ ĥ1,ort ( M1) ĥ3,ort ( M1) ĥp,ort ( M1) ĥp,ort (M2) ]T u(n) = [ Ψ(n+M 1) T Ψ(n+M 1 1) T Ψ(n M 2) T ] T. The learning and adaptation of the coefficients in ĥ ort is done with the LMS algorithm, whose update rule can be written as follows: ĥ ort ĥort + Λy c (n)u(n), (6) where Λ is a diagonal matrix containing the individual stepsizes for each orthogonalized nonlinear basis function. Performing the cancellation process in (5) with the latest estimate of the coefficient vector and then updating the coefficients as shown in (6), and repeating this in every iteration, results in a computationally lightweight and highly adaptive SI cancellation procedure. IV. RF MEASUREMENT RESULTS In order to evaluate the proposed contiguous carrier aggregation based asymmetric full-duplex solution, it is evaluated with real-life RF measurements. The measurement setup is shown in Fig. 4, while all the important parameters are listed in Table I. The measurements are carried out using a National Instruments PXIe-5645R vector signal transceiver, which is used both as a transmitter and a receiver. The used transmit signal is either a 20 MHz or a 40 MHz LTE waveform, centered at 2.46 GHz, and it is fed through a low-cost Texas Instruments CC2595 PA, which amplifies the signal by approximately 23 db. This particular PA is a commercial chip intended to be used in lowcost battery-powered devices, and thereby it produces significant levels of nonlinear distortion. Utilizing such a low-quality PA ensures that the measurement results are representative of a mobile scale device. The PA output signal is then divided between the RF canceller and the circulator, the latter of which is connected to the shared transmit/receive antenna. The SI is leaking to the receiver both via the circulator and from the antenna reflections. The overall transmit power in the measurements is in the order of 6 8 dbm, which is the highest reachable power level with the utilized hardware. In principle, the proposed algorithm works also with higher transmit powers, and demonstrating this is a potential future work item. The total received signal is then routed from the circulator back to the RF canceller, which performs the analog cancellation utilizing the PA output signal as described in [8] and [9]. Finally, the processed signal is fed to the receiver (NI PXIe-5645R) and captured as digital I- and Q-samples, which are post-processed offline to implement digital baseband cancellation. In all the results, the highest nonlinearity order of the digital canceller (P ) is set to 11, and the numbers of pre-cursor (M 1 ) and postcursor taps (M 2 ) are both set to 15. The step-sizes, contained in Λ, are chosen experimentally to provide the best cancellation performance. In the forthcoming results, the LMS-based algorithm is first allowed to converge towards the steady-state coefficient values, after which the cancellation performance is measured in steady-state. This ensures that the results show the true performance of the digital canceller. In the measurements, two different bandwidth scenarios are used: (i) 20 MHz transmit signal with 60 MHz reception bandwidth, and (ii) 40 MHz transmit signal with 120 MHz reception bandwidth. The measurement results from these two scenarios are illustrated in Figs. 5 and 6, respectively. In the figures, the residual SI after the digital cancellers represents the true power of the SI signal from which the noise power has already been excluded. This is done in practice by repeatedly transmitting the same signal and averaging the residual SI over these repetitions. This will obviously reduce the noise while having no effect on the transmit signal dependent SI components. Thanks to this procedure, it is possible to determine the residual SI power even when it is below the receiver noise floor. When investigating the results for the 20/60 MHz case in Fig. 5, it can be observed that the nonlinear digital canceller is indeed capable of attenuating the SI below the receiver noise floor over the whole reception bandwidth. At transmitter passband, the performance is slightly worse, as can be expected, but the power of the residual SI is still approximately at the level of the receiver noise floor. Another intuitive observation is that the linear digital canceller can attenuate the SI only at the own transmitter passband, since it is only capable of modeling the linear coupling characteristics. Hence, it cannot attenuate the SI outside the own transmission bandwidth, and its overall performance is significantly worse than that of the developed nonlinear digital canceller. Figure 6 shows the corresponding results for the case where the transmit signal has a bandwidth of 40 MHz and the reception is done over a bandwidth of 120 MHz. Again, the nonlinear digital canceller is capable of attenuating the residual SI to the level of the receiver noise floor, regardless of the wide reception

5 s s PSD6[dBmNMHz] l2s l4s l6s l8s TX6output6i6746dBmA RX6input6ilf3756dBmA ADC6input6il6f756dBmA Linear6canceller6il76736dBmA Nonlinear6canceller6il86746dBmA RX6noise6floor6il83726dBmA PSD6[dBmNMHz] l2s l4s l6s l8s TX6output6.86dBmA RX6input6.lf49f6dBmA ADC6input6.l58936dBmA Linear6canceller6.l699f6dBmA Nonlinear6canceller6.l8s956dBmA RX6noise6floor6.l8s926dBmA lfss lfss lf2s l5s l4s l3s l2s lfs s fs 2s 3s 4s 5s Frequency6iMHzA lf2s l5s l4s l3s l2s lfs s fs 2s 3s 4s 5s Frequency6.MHzA Fig. 5. The signal spectra after the different cancellation stages, when the bandwidth of the transmit signal is 20 MHz and SI cancellation is done over a 60 MHz bandwidth. bandwidth. Also in this case, the difference to the linear canceller is over 10 db at transmitter passband as well as at neighboring carriers, demonstrating the necessity of nonlinear modeling in a mobile scale device. Overall, these findings regarding the total integrated SI cancellation performance show that the considered method for asymmetric UL and communication is indeed feasible. As long as the nonlinearity of the PA is properly considered in the digital canceller, both the linear and nonlinear SI can be attenuated over significantly wide bandwidths. Regarding the oversampling of the transmit signal, the measurements indicate that under these conditions a sufficiently high cancellation performance is achieved when the nonlinear basis functions are generated with a sampling frequency that is twice the reception bandwidth. Hence, for the 60 MHz reception bandwidth, a sampling frequency of 120 MHz is sufficient, while the 120 MHz reception bandwidth requires a sampling frequency of 240 MHz for the basis function generation. When calculating the required sampling rates with (3), it can be deduced that now, in both of these cases, the 11th-order basis function is partially aliasing on to the reception bandwidth. However, since the higher-order distortions typically tend to be weaker in power, this does not affect the final residual SI. V. CONCLUSION This paper addressed digital self-interference cancellation challenges in a downlink carrier aggregation based mobile fullduplex device where the uplink/downlink rates and bandwidths are asymmetric. The scheme was considered especially in the context of a cellular network, where a full-duplex UE is communicating with a full-duplex capable BS such that the downlink signal bandwidth is wider than the uplink bandwidth. Due to the inherent RF impairments in the UE, nonlinear modeling of the self-interference is required in the digital domain to fully cancel it over the whole reception bandwidth. The feasibility of the proposed scheme was demonstrated with real-life RF measurements using two different bandwidths. In both of these cases, the SI was attenuated below the noise floor over the whole reception bandwidth. Fig. 6. The signal spectra after the different cancellation stages, when the bandwidth of the transmit signal is 40 MHz and SI cancellation is done over a 120 MHz bandwidth. ACKNOWLEDGMENT The research work leading to these results was funded by the Academy of Finland (under the project #259915), and the Linz Center of Mechatronics (LCM) in the framework of the Austrian COMET-K2 programme. The research was also supported by the Internet of Things program of DIGILE, funded by Tekes. REFERENCES [1] M. Heino, D. Korpi, T. Huusari, E. Antonio-Rodríguez, S. Venkatasubramanian, T. Riihonen, L. Anttila, C. Icheln, K. Haneda, R. Wichman, and M. Valkama, Recent advances in antenna design and interference cancellation algorithms for in-band full-duplex relays, IEEE Communications Magazine, vol. 53, no. 5, pp , May [2] D. Korpi, Y.-S. Choi, T. Huusari, S. Anttila, L. Talwar, and M. Valkama, Adaptive nonlinear digital self-interference cancellation for mobile inband full-duplex radio: algorithms and RF measurements, in Proc. IEEE Global Communications Conference (GLOBECOM), Dec [3] M. Duarte, C. Dick, and A. Sabharwal, Experiment-driven characterization of full-duplex wireless systems, IEEE Transactions on Wireless Communications, vol. 11, no. 12, pp , Dec [4] M. Jain, J. I. Choi, T. Kim, D. Bharadia, S. Seth, K. Srinivasan, P. Levis, S. Katti, and P. Sinha, Practical, real-time, full duplex wireless, in Proc. 17th Annual International Conference on Mobile computing and Networking, Sep. 2011, pp [5] D. Bharadia, E. McMilin, and S. Katti, Full duplex radios, in Proc. SIGCOMM 13, Aug. 2013, pp [6] Nokia Solutions and Networks, TD-LTE frame configuration primer, Nov. 2013, white paper. [7] L. Anttila, D. Korpi, V. Syrjälä, and M. Valkama, Cancellation of power amplifier induced nonlinear self-interference in full-duplex transceivers, in Proc. 47th Asilomar Conference on Signals, Systems and Computers, Nov. 2013, pp [8] T. Huusari, Y.-S. Choi, P. Liikkanen, D. Korpi, S. Talwar, and M. Valkama, Wideband self-adaptive RF cancellation circuit for full-duplex radio: Operating principle and measurements, in Proc. IEEE 81st Vehicular Technology Conference (VTC Spring), May [9] Y.-S. Choi and H. Shirani-Mehr, Simultaneous transmission and reception: Algorithm, design and system level performance, IEEE Transactions on Wireless Communications, vol. 12, no. 12, pp , Dec [10] E. Ahmed, A. M. Eltawil, and A. Sabharwal, Self-interference cancellation with nonlinear distortion suppression for full-duplex systems, in Proc. 47th Asilomar Conference on Signals, Systems and Computers, Nov. 2013, pp