REFERENCE RECEIVER ENABLED DIGITAL CANCELLATION OF NONLINEAR OUT-OF-BAND BLOCKER DISTORTION IN WIDEBAND RECEIVERS

Transcription

1 REFERENCE RECEIVER ENABLED DIGITAL CANCELLATION OF NONLINEAR OUT-OF-BAND BLOCKER DISTORTION IN WIDEBAND RECEIVERS Jaakko Marttila, Markus Allén, Marko Kosunen, Kari Stadius, Jussi Ryynänen and Mikko Valkama Department of Electronics and Communications Engineering, Tampere University of Technology P.O. Box 692, FI Tampere, Finland {jaakko.marttila, markus.allen, Department of Micro- and Nanosciences, Aalto University School of Electrical Engineering P.O. Box 13000, FI Espoo, Finland {marko.kosunen, kari.stadius, ABSTRACT This paper proposes digital cancellation of nonlinear distortion originating from out-of-band blocking carriers and induced by the analog front-end nonlinearities in direct-conversion receivers (RXs). The cancellation is enabled by employing an additional reference RX for capturing these blockers. In addition, cancellation of mirrorfrequency interference is targeted. The feed-forward cancellation of nonlinear distortion is blindly adaptive without any a priori information of the received signal or the nonlinearity characteristics of the RX. The performance of this approach is demonstrated with laboratory RF measurements using state-of-the-art base-station hardware. With the distortion cancellation, blocker tolerance of the RX is improved by 7 8 db. 1. INTRODUCTION The popularity of direct-conversion receivers (Rs) has recently increased in both base-stations (BSs) and mobile devices. Compared to, e.g, superheterodyne receiver (RX), the R is more flexible, cost efficient and easier to integrate [1, 2]. However, with increasing number of connected devices, and especially RXs via the introduction of multi-antenna techniques, the cost and size of a single R should be as low as possible. At the same time, modern integrated circuit processes are going towards lower operating voltages. The combination of these factors makes achieving high linearity demanded in, e.g., multi-carrier reception quite a challenge [3]. This problem can, however be alleviated with the means of digital signal processing (DSP). In cellular networks, it is a common scenario that downlink (DL) and uplink (UL) transmission bands in frequency division duplexing are close to each other. This trend is expected to continue in the future 5G systems, in terms of flexible duplexing and associated linearity challenges, especially at carrier frequencies below6ghz [4]. As an example, the DL and UL bands of existing LTE networks are often separated by couples of tens of megahertzes. At the same time, the width of these bands can be 60MHz or even more. In such scenario, already strong third order nonlinear distortion can fall from band to another. For example an UL BS RX can suffer from nonlinear distortion caused by co-located DL transmission entering the low-noise amplifier () and the distortion falling inside the UL band. Such DL carriers are referred to, herein, as out-of-band (OOB) blockers. In the scientific literature, multiple approaches for linearizing a R by the means of DSP have been proposed [5 7]. However, these approaches are limited in canceling the nonlinear distortion stemming from the carriers inside the reception band of the RX to be linearized. At the same time, analog nonlinearity modeling has been proposed in [8,9]. However, this results in inaccuracies in modeling and needs multiple, tailored RX front-ends together with the nonlinear analog circuits used for modeling the nonidealities. In this paper, digital cancellation of distortion originating from OOB blockers with the help of a reference receiver (ref-rx) is proposed. To the best of authors knowledge, this is the first time fully digital modeling and cancellation of the OOB blocker distortion is demonstrated. The ref-rx allows observing also OOB blockers and thus enables digital cancellation of this distortion. The cancellation principle is based on combining the observations of the main RX and the ref-rx and thereafter regenerating an estimate of the distortion impairing the reception of the main RX. Thereafter, this estimate is used to cancel the main RX distortion with the means of adaptive filtering. The cost of this linearization approach is the ref-rx itself. However, e.g., in carrier aggregation scenarios, the cost could be only in the internal digital front-end communications, if inherently present RXs can be exploited. The rest of the paper is organized as follows. Section 2 discusses modeling of the nonlinearities in Rs. Section 3 introduces the out-of-band adaptive interference cancellation (OOB-AIC) solution proposed herein. The RF measurement results illustrating the linearization performance are given in Section 4. Finally, the paper is concluded in Section NONLINEARITY MODELING IN DIRECT-CONVERSION RECEIVERS Herein, a parallel Hammerstein nonlinearity model [10] is applied for modeling the RX nonlinearities highlighted in the conceptual block diagram of a R in Fig. 1, together with the main sources of mismatches. Separate models are used for the RF and baseband (BB) components. In addition, mismatch of RX components is considered in the modeling as well as in the compensation approach. The joint effects of cascaded nonlinearities and mismatch are analyzed more thoroughly, e.g., in [5, 7]. Mathematically, the odd RF polynomial orders, considered up tok RF, are gathered intoω RF, i.e.,ω RF = {1,3,...,K RF}. With this assumption, the output as a discrete-time baseband-equivalent

2 RF Front-End x y 0 90 LO Baseband Processing z=z + jz I z I Q z Q Digital and Linearization Chan. Eq. Detection Decoding Synch. Table 1. Most Important Terms Generated by the Joint Nonlinearity Model in Case of Ω RF = {1,3} and Ω BB = {1,2,3}. Terms Mirror Terms Interpretation of the Term x(n) x (n) Original signal aroundf IF x(n) 2-2nd-order IMD around BB Fig. 1. A conceptional block diagram of a wideband R. The main sources of nonlinear distortion and mismatch are highlighted. Time index t is omitted from signal notation for illustration convenience. model can be expressed as ( ) 1 k y(n) = a k (n) 2 k 1 k+1 [x(n)x (n)] (k 1)/2 x(n) 2 k Ω RF = ã k (n) x(n) k 1 x(n), k Ω RF wherea k (n) are the impulse-responses for each odd polynomial order inω RF taking memory effects into account andx(n) is the original input [7, 11]. For Ω RF, one practical example is third-order distortion modeling, Ω RF = {1,3}, where 1 represents the linear component. This signal y(n) is fed into the mixers, BB amplifiers, filters and analog-to-digital converters (s). The observed digital waveform is z(n) = c re,k (n) y k I (n)+jc im,k (n) yq(n) k k Ω BB = k Ω BB l=0 k c k,l (n) y k l (n)[y (n)] l, wherec re,k (n) andc im,k (n) are the impulse responses for each polynomial order in Ω BB for I and Q rails, respectively, considering also memory effects [7, 11]. In addition, c k,l (n) are the modified filters reflecting the combined responses for each of the complex-valued signal components. Also the mismatch and nonlinearity of the mixers is included in (2). For example, the mismatch appears as a conjugate response for the linear component. Applying the scenario ofω RF = {1,3} andω BB = {1,2,3}, the most important distortion components are presented in Table 1. Also higher order components are present [5, 7], but in practical scenarios they rarely appear above the measurement noise floor [5]. 3. PROPOSED ADAPTIVE CANCELLATION OF OUT-OF-BAND BLOCKER DISTORTION Herein, the operation principle of the proposed OOB-AIC is described and combined with circularity restoring mirror-frequency interference (MFI) cancellation. The MFI cancellation is performed first for both the digitized signals and thus also discussed here shortly before the linearization process. This is illustrated with a conceptional block diagram in Fig Mirror-Frequency Interference Cancellation The main RX and the ref-rx correction stages shown in Fig. 2 aim in restoring circularity of the received complex-valued signals (1) (2) x 2 (n), [x (n)] 2-2nd-order harmonics around±2f IF x(n) 2 x(n) x(n) 2 x (n) 3rd-order IMD around f IF [x (n)] 3 x 3 (n) 3rd-order harmonics around 3f IF x(n) 2 x 2 (n), x(n) 2 [x (n)] 2 - x(n) 4 x(n) 4th-order IMD around±2f IF x(n) 4 x (n) 5th-order IMD aroundf IF x(n) 2 [x (n)] 3 x(n) 2 x 3 (n) 5th-order IMD around 3f IF adaptively [12]. The correction concentrates on cancelling the MFI Mirror Terms of Table 1 induced by the mismatches of the RX mixers and BB components. The performance of the MFI cancellation is demonstrated in Section 4 together with the actual linearization results and Reference RX Signal Combining After the correction, the linearization stage, highlighted with red color in Fig. 2, aims in canceling the remaining spurious Terms in Table 1. This starts with combining the main RX and the ref-rx signals, which is shown in the conceptional block diagram of Fig. 2, where the ref-rx signal is shifted in frequency in order to mimic the original composite RF waveform. In order to facilitate this, the sample rate of both the signals need to be high enough to avoid aliasing. This higher sample rate depends on the exact reception scenario at hand. The shifting frequency f shift is the difference of the main RX and the ref-rx down-conversion local oscillator frequencies and thusf shift = f LO,1 f LO,2. Overall, the combined signal is written as z comb(n) = z main(n)+z ref(n)e j2πfshiftn. The combining process is demonstrated with a practical measurement example using the four-carrier RF waveform shown in Fig. 3. There, the main RX and ref-rx bands are denoted with red lines. The carrier setup corresponds to the LTE band 3. Three carriers captured by the main RX lie on the UL band and the additional ref-rx band carrier is assumed to arrive from a nearby DL BS transmitter. Spectra of the digitized waveforms are shown in the top plots of Fig. 4. Thereafter, the ref-rx signal is frequency-shifted to the digital center-frequency that corresponds to the position of these bands on the RF for regeneration of the distortion components originating from the composite RF signal. In this example, the frequency shift enables modeling of the intermodulation between the two strong carriers shown in Fig. 3 and especially the distortion components falling inside the main RX UL band Nonlinear Distortion Estimation and Cancellation Herein, the nonlinear distortion cancellation first proposed in [6] is extended to cover also distortion originating from OOB blockers by employing the ref-rx and signal combining process described in Subsection 3.2. The complete signal flow is illustrated in Fig. 2. The learning stage separates nonlinear distortion outside the actual bandstop filtered blocker bands. This filtering is performed in order to avoid biased solutions due to the correlation of, e.g.,

3 TX DPLX FE with f LO,1 z main z main z comb BSF BPF z BP f 1 () Delay BSF z BS w 1 Learning ˆd - ˆx BS Ref-RX FE with f LO,2 Ref. z ref j2 fshiftn e z ref Combining Delay d w 1 - Cancellation ˆx High sample rate sections Fig. 2. Block diagram illustrating the principle of the proposed OOB-AIC method for main RX correction and linearization. The MFI suppression is implemented per RX, whereafter the nonlinearity cancellation stage linearizes the main RX band. The correction stages are implemented based on [12]. Time index n is omitted from signal notation for illustration convenience Band Ref-RX Band Fig. 3. The RF signal setup for BS reception with the main RX and the ref-rx. The main RX captures the three 10MHz carriers located on the LTE Band 3 UL duplexing band (1710MHz to 1785MHz). The ref-rx captures the single10mhz carrier located on the LTE band 3 DL duplexing band (1805MHz to 1880MHz). The weak carrier powers are 67dBm and the strong carrier powers 27dBm at the duplexer output (splitter input). third-order distortion and the linear component [6] and is realized by the bandpass and bandstop filters (BPF and BSFs). The main RX band filters do suppress the out-of-band contribution of the reference nonlinearities and thus allow learning proper coefficients for canceling the inband distortion. In parallel, the cancellation stage uses the whole, unfiltered distortion estimate for improved performance on the blocker bands and the neighboring bands. Also in here, the main RX band filter suppresses out-of-band signal content. Formally, the nonlinearity cancellation in the proposed OOB-AIC is performed so that ˆx(n) = z comb(n) w H (n)d(n), (3) where ˆx(n) is the final linearized signal,w(n) contains the impulse responses of the adaptive filters (AFs), and d(n) contains the main RX band contribution of the considered reference nonlinearities f 1( ),f 2( ),,...,f P( ), chosen, e.g., from Table 1, without bandstop filtering. More specifically, the main RX band filtered samples of P reference nonlinearities are stacked into a single vector where d(n) = [d 1(n),d 2(n),...,d P(n)] T, (4) d 1(n) = h main(n) f 1(z BP(n)), d 2(n) = h main(n) f 2(z BP(n)), (5) Fig. 4. Top: the observed signal spectra are shown separately for the main RX and the ref-rx. Bottom: combined signal spectra is shown, illustrating the increased sample rate and frequency shift for the ref-rx observation. and so on. The bandstop filtered distortion estimates are denoted as d BS(n) = [d BS,1(n),d BS,2(n),...,d BS,P(n)] T. (6) Also the AF coefficients are gathered into a single vector w(n) = [w 1(n),w 2(n),...,w P(n)] T. (7) If no a priori information is available for the least-mean-square learning [13], the AF vector w is initialized as w(0) = 0 PM 1, where M is the length of the individual AFs. Thereafter, the combined AF output for n = 0,1,2,... is This results in the learning error and finally the AF update is given by ˆd(n) = w H (n)d BS(n). (8) ˆx BS(n) = z BS(n) ˆd(n) (9) w(n+1) = w(n)+diag(µ)ˆx BS(n)d BS(n), (10)

4 VECTOR SIGNAL GENERATOR LPF RX1 RX2 DUAL RX EVB FPGA Clock MATLAB Fig. 5. The measurement setup where the devices are controlled by a single PC running Matlab. where diag( ) denotes a function for converting a vector to a diagonal matrix. In OOB-AIC, these AF coefficients w are used in the actual cancellation as described in (3). 4. MEASURED LINEARIZATION PERFORMANCE In this section, the performance of the proposed OOB-AIC is demonstrated with RF measurements using true BS RX hardware. The measurement setup is illustrated in Fig. 5. The reference nonlinearities applied in the cancellation processing are the ones shown in Table 1. The learning is performed as least-squares block solution [13]. The signal scenario used in the measurements is the same as discussed in Section 3. An example of the original RF spectum is shown in Fig. 3. Four LTE carriers with 16-QAM data-modulation are generated at the RF corresponding to the LTE band 3 DL and UL duplexing bands. Two of these carriers are weak compared to the other two, the latter ones being the main source of nonlinear distortion. One of the strong carriers is located inside the UL band and the other one on the DL band, mimicing a nearby interfering BS transmitter. This signal is split to the main RX (f LO,1 = 1745MHz and 100MHz bandwidth) receiving the UL band and to the ref-rx (f LO,2 = 1845MHz and 100MHz bandwidth) tuned for the DL band. In the main RX chain, a HD Communications Corp. HD24089 wideband is used. Thereafter a state-of-the-art dual RX evaluation board is used to down-convert, digitize and capture the signals. The resulting signal-to-noise-and-distortion ratio (SNDR) and symbol error ratio (SER) numbers are evaluated and averaged over ten independent measurements and data realizations of samples. An example of the received main RX signal is shown in Fig. 6 together with the linearized result. In the figure, it is visible that nonlinear distortion appears around the strong carrier (@35 MHz) and around the other weak carrier (5 MHz). This distortion around 5 MHz results from intermodulation between the inband blocking carrier and the OOB blocking carrier and thus for its cancellation it is essential to capture also the OOB blocker. Furthermore, MFI originating from the inband blocker is visible around 35 MHz. This MFI is pushed below the noise floor by the main RX correction. In Fig. 7, the SNDR and SER for the two weak carriers are plotted. For example, with 27dBm blocker powers, the SNDR of the weak carriers around5mhz and25mhz are improved from13db to 19dB and from 11dB to 20dB, respectively. Alternatively, the level of 14dB SNDR can be maintained with 7dB and 8dB higher blocker powers compared to the scenario without linearization. Similar improvements are visible also in the SER results plotted in the bottom of the Fig. 7. For example, 1% SER level is maintained with7db and8db higher blocker powers compared to the scenario without linearization. With 27 dbm blocker powers, the SERs of the5mhz and25mhz carriers are improved from13% and5% to 0.01% and 0.03%, respectively. Received () Linearized Fig. 6. An example of the the original received main RX spectrum together with the final linearized spectrum. The weak desired carriers are located around 5MHz and 25MHz. SNDR [db] SER Received, MHz Linearized, MHz Received, MHz Linearized, MHz Received Power per Blocker [dbm] 10 0 Received, MHz 10-1 Linearized, MHz Received, MHz 10-2 Linearized, MHz Received Power per Blocker [dbm] Fig. 7. The SNDR (top) and SER (bottom) of the weak LTE carriers around5mhz and25mhz down-converted center-frequencies as a function of the blocker carrier power. The power of the weak carriers is constant 67dBm. 5. CONCLUSION In this papers, digital cancellation of nonlinear OOB blocker distortion was proposed. This addresses a significant weakness in numerous state-of-the-art digital linearization approaches reported in the scientific literature. The proposed OOB-AIC combines the main RX observation with the OOB blockers captured with a ref-rx and thereafter regenerates an estimate of nonlinear distortion induced by the RX analog front-end components. This estimate is used to cancel the distortion present in the main RX observation, effectively linearizing this RX. The performance of the proposed method was illustrated with RF hardware measurements using modern LTE waveforms and true frequency allocations. This improvement in RX blocker tolerance allows more flexibility in the analog component design and increased reception dynamic range. This is expected to be critical in the emerging 5G systems, with increased requirements for flexible duplexing and RF spectrum use, especially at carriers below 6GHz.

5 6. REFERENCES [1] A. A. Abidi, Direct-conversion radio transceivers for digital communications, IEEE J. Solid-State Circuits, vol. 30, no. 12, pp , Dec [2] B. Razavi, Design considerations for direct-conversion receivers, IEEE Trans. Circuits Syst. II, Analog Digit. Signal Process., vol. 44, no. 6, pp , June [3] B. Razavi, Cognitive radio design challenges and techniques, IEEE J. Solid-State Circuits, vol. 45, no. 8, pp , Aug [4] P. Pirinen, Challenges and possibilities for flexible duplexing in 5G networks, in Proc. IEEE 20th Int. Workshop Comput.- Aided Modeling, Anal. and Design of Commun. Links and Networks, Sep. 2015, pp [5] M. Grimm, M. Allén, J. Marttila, M. Valkama, and R. Thom, Joint mitigation of nonlinear RF and baseband distortions in wideband direct-conversion receivers, IEEE Trans. Microw. Theory Tech., vol. 62, no. 1, pp , Jan [6] M. Allén, J. Marttila, M. Valkama, S. Singh, M. Epp, and W. Schlecker, Digital full-band linearization of wideband direct-conversion receiver for radar and communications applications, in Proc. 49th Asilomar Conference on Signals, Systems and Computers, Pacific Grove, CA, Nov. 2015, pp [7] J. Marttila, M. Allén, M. Valkama, M. Kosunen, K. Stadius, and J. Ryynänen, Reference receiver enhanced digital linearization of wideband direct-conversion receivers, IEEE Trans. Microw. Theory Tech., 2016, submitted, revised. [8] E. Keehr and A. Hajimiri, Equalization of third-order intermodulation products in wideband direct conversion receivers, IEEE J. Solid-State Circuits, vol. 43, no. 12, pp , Dec [9] E. A. Keehr and A. Hajimiri, Successive regeneration and adaptive cancellation of higher order intermodulation products in RF receivers, IEEE Trans. Microw. Theory Tech., vol. 59, no. 5, pp , May [10] D. R. Morgan, Z. Ma, J. Kim, M. G. Zierdt, and J. Pastalan, A generalized memory polynomial model for digital predistortion of RF power amplifiers, IEEE Trans. Signal Process., vol. 54, no. 10, pp , Oct [11] H. Gutierrez, K. Gard, and M.B. Steer, Nonlinear gain compression in microwave amplifiers using generalized powerseries analysis and transformation of input statistics, IEEE Trans. Microw. Theory Tech., vol. 48, no. 10, pp , Oct [12] L. Anttila, M. Valkama, and M. Renfors, Circularity based imbalance compensation in wideband direct-conversion receivers, IEEE Trans. Veh. Technol., vol. 57, no. 4, pp , July [13] S. Haykin, Adaptive Filter Theory, Prentice Hall, Upper Saddle river, NJ, 4th edition, 2002.