Implementation of a 2 by 2 MIMO-GFDM Transceiver for Robust 5G Networks

Implementation of a 2 by 2 MIMO-GFDM Transceiver for Robust 5G Networks Martin Danneberg, Nicola Michailow, Ivan Gaspar, Maximilian Matthé, Dan Zhang, Luciano Leonel Mendes, Gerhard Fettweis Vodafone Chair Mobile Communication Systems, Technische Universität Dresden, Germany Inatel, Sta. Rita do Sapucai, MG, BR {martin.danneberg nicola.michailow ivan.gaspar maximilian.matthe dan.zhang fettweis} @ifn.et.tu-dresden.de, luciano@inatel.br Abstract The fifth generation (5G) of mobile cellular systems will demand an unprecedented flexibility of the physical layer (PHY). Several approaches are being proposed based on theoretical and simulation analyses, but there is a lack of implementations that allow for performance evaluation under real channel conditions and front-end impairments. In this paper we show that that a 2x2 multiple-input multiple-output (MIMO) Generalized Frequency Division Multiplexing (GFDM) transceiver can be implemented using the new National Instruments USRP-RIO, which proves that today s technology is ready to afford the complexity of modern waveforms. In this implementation we explore the flexibility of the USRP combined with LabVIEW, allowing for an easy integration between software host processing and real-time processing based on Field Programmable Gate Arrays (FPGAs). This implementation allows to evaluate different GFDM characteristics, such as reduced latency, low out-ofband (OOB) emission and increased robustness in real-world environment. Index Terms 5G, GFDM, Channel Estimation, MIMO, USRP I. INTRODUCTION Apart from ever growing demand for higher data rates, future 5G wireless communication systems need to deliver many other services, such as Tactile Internet, massive machine communication and dynamic spectrum usage [1], [2]. The well-known Orthogonal Frequency Division Multiplexing (OFDM) has been a widely adopted waveform in fourth generation (4G) systems. However, it has shown difficulty addressing upcoming new types of services. Therefore, it is expected that new waveforms will be dominant in 5G systems [3]. Driven by this vision, many different waveform proposals tailored for individual 5G scenarios have appeared in the literature, e.g., filter-bank multi-carrier (FBMC) [4], universal-filtered multi-carrier (UFMC) [5] and Faster than Nyquist (FTN) signalling [6]. While we can select one specific waveform for each 5G application, it is more desirable to adopt a single, flexible waveform that can be easily configured to address a multitude of scenarios. The introduction of such a waveform will lead to a disruptive change in the design of 5G systems. In this context, GFDM [7] is a promising candidate to enable softwaredefined waveform engineering. First, its flexible characteristics can be easily tuned to address the new requirements. For instance, GFDM adopts a circular pulse shaping filter, which means that the block structure does not present filter tails, favoring latency constrained applications. By choosing a pulse shaping filter with good frequency localization, the achieved low OOB emission is beneficial to dynamic spectrum access and carrier aggregation [8], [9]. As a non-orthogonal waveform, GFDM can tolerate loose time and frequency synchronization [10], which is a important feature for device to device communication. Next, GFDM provides a very flexible timefrequency structure such that prominent waveform candidates for 5G and legacy waveform, i.e. OFDM and Single Carrier Frequency Domain Equalization (SC-FDE), can be emulated as corner cases, making GFDM an interesting solution for accomplishing a smooth migration from 4G to 5G. From theory to practice, the development of a proof-ofconcept GFDM testbed is an important step. This paper contributes to the development of a 2 by 2 MIMO GFDM testbed based on National Instruments (NI) universal software radio peripheral with reconfigurable input/output (USRP-RIO) platform. The details of testbed development include the employment of time-reverse space time coding TR-STC for achieving diversity gain, and preamble-based synchronization and channel estimation. Relying on this proof-of-concept testbed, we can demonstrate that MIMO-GFDM has an affordable complexity. In future, we can optimize its configuration and validate its use for 5G in close to reality. The remainder of the paper is organized as follows. Section II explains the GFDM and the theory behind the transceiver. A overview about the implementation and the hardware platform is given in Section III. The results from both simulation and the demonstrator are presented in Section IV. The paper is concluded in Section V. II. MIMO-GFDM SYSTEM MODEL In this section, we present the GFDM principles and show how space time code (STC) can be applied to this nonorthogonal waveform. Further the implemented synchronization and channel estimation algorithms are described. A. GFDM Background GFDM is a multicarrier modulation scheme where K subcarriers and M subsymbols are used to transmit N = KM complex data symbols d k,m, k = 0,..., K 1, m = 0,..., M 1 from a given constellation set, e.g. 16QAM. A prototype filter g[n], n = 0,..., N 1 is circularly shifted

in the frequency domain to provide the pulse shape for each subcarrier. Circular convolution between the data sequence and the K subcarriers is used to generate the GFDM transmit samples x[n] as x[n] = = m=0 m=0 ( d k,m δ[n mk] g[n] exp (j2π kk )) n g [ n mk N ] (1) d k,m exp (j2π kk ) n, (2) where N denotes modulo N. The second summation in (2) is the K-point Inverse Discrete Fourier Transform (IDFT) of the data symbols transmitted in the mth subsymbol, but concatenated M times because n = 0, 1,..., N 1. Let d m [n] = K { } K d( ),m n = d k,m exp (j2π kk ) n, (3) where K {( )} n is the K-point inverse discrete Fourier transform (DFT) evaluated at time n. Hence, the GFDM signal can be rewritten as x[n] = m=0 d m [n]g [ n mk N ]. (4) Before the signal is transmitted through the channel, a cyclic prefix (CP) is added, leading to x and a window u multiplies the entire block. Assuming a flat noiseless channel, i.e., y[n] = x[n], a receiver filter γ[n] derived from g[n] [11] can be used to recover the information after the CP is removed, leading to ˆd k,m = ( γ [ n N ] y[n] exp ( j2π k K n)) n=mk = N 1 n=0 γ [ n mk N ]y[n] exp( j2π km KM n) = F N {γ [ n mk N ]y[n]} km. Eq. (5) shows that the N-point DFT is only evaluated at every Mth sample. Also, notice that (5) holds for matched filter (MF), zero-forcing (ZF) or minimum mean square error (MMSE) receivers and it was used to build the low complexity GFDM demodulator employed in the demonstrator. B. TR-STC applied to GFDM The time-reverse space time code (TR-STC) [12] was first proposed as a STC method for single carrier systems over frequency-selective channels (FSCs). Since the block length of the GFDM signal is limited to N samples, GFDM and TR- STC can be easily integrated, achieving a flexible and robust multicarrier waveform that can explore full diversity gain. The main idea of the TR-STC-GFDM is to apply the space-time coding on the samples of the GFDM signal in the frequency-domain. In order to achieve this goal, two successive GFDM blocks x i [n] and x i+1 [n] are generated based on two independent sets of data symbols d (i) k,m and (5) d (i+1) k,m, respectively. STC is applied to the samples of the GFDM blocks according to Block i Block i + 1 Tx Antenna 1 Tx Antenna 2 N {X i[f]} { N X i+1 [f] } (6) N {X i+1[f]} N {X i [f]}, where X ( ) [f] = F N { x( ) } and i is an even number. The name time-reversal derives from the following property of the IDFT: N {X i [f]} = x i [ n N ]. (7) After synchronization and removal of the CP, the signals in the frequency-domain at the lth receive antenna, assuming a time-variant multipath channel with channel impulse response (CIR) h j,l [n] between the jth transmit and lth receive antenna, are given by Y i,l [f] = H 1,l [f]x i [f] H 2,l [f]x i+1[f] + W 1,l [f] (8) Y i+1,l [f] = H 1,l [f]x i+1 [f] + H 2,l [f]x i [f] + W 2,l [f], (9) where H j,l [f] = F N {h j,l [n]} is the channel frequency response and W i,l [f] is the additive white Gaussian noise (AWGN) in the frequency-domain for the GFDM block at the lth receive antenna. The received signals can be combined as follows to achieve full diversity gain: where ˆX i [f] = H 1 eq [f] ˆX i+1 [f] = H 1 eq [f] L l=1 L l=1 H eq [f] = H 1,l[f]Y i,l [f] + H 2,l [f]y i+1,l[f] H 1,l[f]Y i+1,l [f] H 2,l [f]y i,l[f], 2 j=1 l=1 (10) L H j,l [f] 2 (11) and L is the number of receiving antennas. Finally, the estimated received signal that must be demodulated in accordance with (5) is given by ˆx i [n] = N C. Synchronization and Channel Estimation { ˆXi [f]}. (12) For the purpose of synchronization and channel estimation, a known preamble is prepended before the STC data block, leading to the frame structure illustrated in Figure 1. As indicated, the preamble is transmitted with a 3 db higher power. Given a N ch taps long CIR h j,l [n], the received signal is given by y[n] = N ch 1 l=0 h[l] x[n l], (13) r[n] = y[n θ]e j2πɛn + w[n], (14) where θ is the symbol timing offset (STO), ɛ is the carrier frequency offset (CFO) and w[n] denotes the AWGN. The

Antenna 1 preamble x p,1 [n] payload x 1 [n] x 2 [n] N w N CP N w N w N CP N N w Antenna 2 x p,2 [n] x 2[ n N ] x 1[ n N ] Time Fig. 1. The TR-STC GFDM frame structure. d 1 d 2 GFDM mod. GFDM mod. TR-STC preamble LUT x d,1 x d,2 x d,3 x d,4 x p,1 x p,2 x d,1 x d,2 x d,3 x d,4 x p,1 x p,2 assemble packet x 1 x 2 DAC RF RF ADC time synch y 1 y 2 s split packet ỹ d,11 ỹ d,12 ỹ d,21 ỹ d,22 ỹ p,11 ỹ p,12 ỹ p,21 d p,11 d p,12 d p,21 y d,11 y d,12 y d,21 y d,22 channel estimator h11 h12 h21 channel equalizer, resampling, -STC ˆxd,1 ˆxd,2 GFDM demod. GFDM demod. d 1 d 2 ỹ p,22 d p,22 h22 Fig. 2. Transceiver block diagram. symbol duration needs to be chosen such that it is smaller than the coherence time of the channel. Consequently, the CIR varies slower than the frame rate of the transmission. On the receiver side, it is assumed that CFO is null and digital-to-analog converter (DAC) operates with exactly the same sampling rate as the analog-to-digital-converter (ADC) in the transmitter due the use of the same reference clock. The start index of a frame is detected by cross-correlating the received signal on channel 1 r 1 [n] with the known preamble x p,1 [n], leading to the metric C[n] = 1 and selecting ˆθ = argmax n r 1[n + k]x p,1 [k], (15) C[n]. Further, both known preambles are used for channel estimation. The channel coefficients are obtained according to Ĥ j,l [f] = Y p,j,l[f] X p,j,l [f], (16) where Y p,j,l [f] are the samples of the received preamble and X p,j,l [f] are the samples of the transmit preamble from a lookup table (LUT) in frequency domain. Since the preamble can be shorter than the data block, the channel is estimated with a smaller number of samples than required by the equalizer. In this implementation, the resampling is carried out in two steps. First, the frequency responses of the channels Ĥ j,l = (Ĥj,l [f]) are multiplied with a pseudo-inverse of a modified DFT matrix. The DFT matrix is defined as F = 1 ( ) e j2πuv Np, (17) Np N cp where u = 0,..., 1 and v = 0,..., N cp 1. Note that the index v is limited to N cp, because the channel is assumed to be shorter than the CP. Applying F to Ĥ j,l yields the time domain channel impulse responses ) ĥ j,l = (ĥj,l [n] according to ĥ j,l = F + Ĥ j,l for j = 1, 2 and l = 1, 2. (18) In the next step, interpolation is performed with a oint Fast Fourier transform (FFT), which is equivalent to the size of payload, yielding H j,l [f] = F N {ĥj,l[n]} for j = 1, 2 and l = 1, 2. (19) The channel coefficients H j,l [f] can now be used directly in (10). III. IMPLEMENTATION MODEL AND PLATFORM A. Implementation Model An overview of the processing blocks used to implement the transmitter and receiver can be found in Figure 2. In the transmitter, the inputs d 1 and d 2 are vectors that carry N Quadrature Amplitude Modulation (QAM) symbols each. All GFDM-related processing takes place in the subsequent waveform modulator block, i.e., the data symbols are arranged in a time-frequency grid and the subcarrier filter is applied, yielding the modulated waveforms x d,11 and x d,12. Those two signals are fed to the TR-STC block, which applies the spacetime encoding according to (6) in order to obtain x d,21 = x d,12[ n N ] and x d,22 = x d,11[ n N ]. (20) Next, the two predefined preamble signals x p,11 and x p,22 are acquired from a LUT. From this point, all six signals

undergo the same processing steps of cyclic prefix extension and windowing. Lastly, the frame is assembled according to the structure displayed in Figure 1. The samples of the two channels passed to the hardware are collected in x 1 and x 2. On the receiver side, the software-defined radio (SDR) hardware provides the received signals y 1 and y 2, which each contain twice as many samples as in one transmitted frame. The signal on the first channel is used to obtain the start index of the the frame s through cross-correlation with the known preamble. In the subsequent ion blocks, the received signals y 1 and y 2 are split up into four data blocks y d,11, y d,12 and y d,21, y d,22 and four preamble blocks y p,11, y p,21, y p,22 and y p,12, where the first index indicates the receive antenna and the second index corresponds to the time slot. The CP is removed in the same step. The channel estimation block uses the reference preamble signals x p,1, x p,2 to calculate the channel impulse responses h 11, h 12, h 21, h 22 according to (16). In the following block, the channel responses are resampled according to (18) and (19) and then used by the space-time decoder to combine the received samples according to (10). The results are the estimated receive samples ˆx d,1 and ˆx d,2. These are then passed to the GFDM demodulator, which implements a linear receiver, i.e., ZF, to detect the data symbols ˆdd,1 and ˆdd,2. From the constellation points the error vector magnitude (EVM) and instantaneous symbol error rate (SER) used to evaluate the system performance. B. Implementation Platform The described GFDM MIMO System is implemented on the National Instruments SDR platform USRP-RIO 2953R. The main advantage of this platform is the integrated Kintex- 7-FPGA, which allows to host all relevant parts of the signal processing. The other main advantage is the RF front-end with two RF channels. This enables a MIMO GFDM transceiver in one device, without the need of synchronization and data transfer between several front-ends with only one channel. The USRP-RIO has a frequency range of 1.2 to 6 GHz with a bandwidth of 40 MHz. A PC is connected to the USRP via the NI PXIe-PCIe8371 card with a throughput rate of 832 MB/s. In general, the USRP allows easy prototyping of advanced waveforms in a frequency range that covers the main bandwidths used for mobile communications today. Another great feature of this platform is that the communication chain can be partially implemented in software and in hardware. LabVIEW Communications Design Suite is an versatile tool that allows the implementation of software functions in combination with with block designed with hardware description language to explore the flexibility of SDR combined with the speed of FPGA implementation. IV. THE MIMO-GFDM DEMONSTRATOR The demonstrator consists of 2 USRP-RIO terminals, one configured as transmitter and another set as receiver, as presented in Figure 3. Each USRP is connected to a controlling PC. Both USRPs are connected to a reference clock Meinberg GPS Receiver 170MP to eliminate frequency offsets. On the transmitter host PC, data is generated, modulated with the LabVIEW Communications MIMO-GFDM modulator Control PC RF0 USRP-RIO FPGA Tx RF1 reference clock Fig. 3. Schematic overview of the hardware setup. TABLE I PARAMETERS LabVIEW Communications MIMO-GFDM demodulator Control PC RF0 USRP-RIO FPGA Rx RF1 Parameter Variable Value number of subcarriers (data) K, K p 128 number of subcarriers (preamble) K p 128 number of subsymbols (data) M 15 number of subsymbols (preamble) M p 5 number of transmit samples (data) N = KM 1920 number of transmit samples (preamble) = K pm p 1920 set of active subcarriers (data) K set 11 to 54 set of active subcarriers (preamble) K p,set 10 to 55 set of active subsymbols (data) M set 1 to 15 set of active subsymbols (preamble) M p,set 1 to 5 prototype subcarrier filter g RRC prototype filter roll-off α 0.4767 window coefficients w window function roll-off β 0.5 length of window flanks N w = βk 32 maximum channel delay spread N ch length of cyclic prefix N CP = N ch 32 Modulation order M u 4 number of channels - 2 center frequency - 1.99 GHz sampling frequency - 10 MHz transmitter gain - 0 db receiver gain - 0 db GFDM waveform and TR-STC encoded. Table I presents the relevant parameters used in this demonstrator. The digital in phase and quadrature (IQ) samples are transferred via the PCI- Express connection to the FPGA on the transmitter USRP, which then performs digital to analog conversion and send the signal over the air. On the receiver side, the USRP captures IQ samples and passes them to the receiver host PC via the PCI-Express connection, where the waveform is demodulated. Figure 4 shows the measured spectrum at a Rhode&Schwarz FSQ8 spectrum analyzer and compares it with simulation results, while Figure 5 present the graphical user interface (GUI) for the transmitter and receiver, respectively, showing the transmitted waveforms in frequency and time domains and the received spectrum, estimation of the channels and quality of the demodulated constellation. The measured OOB radiation is at around -48 db. This result is achieved due to the pulse shaping filter applied at each subcarrier. Therefore, GFDM is favorable in communication scenarios with highly fragmented spectrum or in Cognitive Radios (CRs). In [13], the authors show that GFDM has less influence and disturbance to legacy systems than OFDM. The simulated combined OOB

Receive power [dbm] 60 80 100 120 140 measured 160 simulated 180 1.990 1.992 1.994 1.996 1.998 2 Frequency [GHz] implementation of a GFDM based 2 2 MIMO system on the NI USRP-RIO platform. The developed demonstrator is used for proof-of-concept of GFDM as a candidate waveform for future wireless communication systems. As a part of future works, the developed MIMO-GFDM demonstrator will be integrated into the NI LTE Application Framework. This solution will assist further exploration on the flexibility of GFDM in respect to resource management and scheduling algorithms utilized by the MAC layer. Fig. 4. Comparison of measured and simulated spectrum. VI. ACKNOWLEDGEMENTS This work has been performed in the framework of ICT-318555 5GNOW which is partly funded by the European Union. The authors thank Instituto Nacional de Telecomunicações (Inatel) and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq Brasil) for partially funding the work presented in this paper. The authors also thanks National Instruments for supplying all necessary software and hardware tools to implement the demonstrator and for providing unconditional technical support. REFERENCES (a) Screenshot of the TX-GUI (b) Screenshot of the RX-GUI Fig. 5. Screenshots of the control panel designed with LabVIEW Communications Design Suite. level is considerably lower than the measured OOB level because it does not take into account the spectrum analyzer sensitivity and RF impairments, such as IQ-imbalance and nonlinearities, introduced by the USRP front-end. V. CONCLUSION Due to its flexible time-frequency structure, GFDM has been researched to serve diverse scenarios envisioned for 5G. Particularly, its integration with MIMO is one key research topic, which includes the design of the TR-STC for achieving diversity gain and algorithms for synchronization and channel estimation. In order to evaluate and validate the research results in real environments, this paper presents in detail the [1] G. P. Fettweis, The Tactile Internet: Applications and Challenges, IEEE Vehicular Technology Magazine, vol. 9, no. 1, pp. 64 70, Mar. 2014. [2] F. Boccardi, R. Heath, A. Lozano, T. Marzetta, and P. Popovski, Five disruptive technology directions for 5G, IEEE Communications Magazine, vol. 52, no. 2, pp. 74 80, Feb. 2014. [3] G. Wunder et. al., 5GNOW: non-orthogonal, asynchronous waveforms for future mobile applications, IEEE Communications Magazine, vol. 52, no. 2, pp. 97 105, Feb. 2014. [4] R. Zakaria and D. Le Ruyet, A novel filter-bank multicarrier scheme to mitigate the intrinsic interference: application to MIMO systems, IEEE Transactions on Wireless Communications, vol. 11, no. 3, pp. 1112 1123, Mar. 2012. [5] V. Vakilian, T. Wild, F. Schaich, S. ten Brink, and J.-F. Frigon, Universal-Filtered Multi-Carrier Technique for Wireless Systems Beyond LTE, in 9th International Workshop on Broadband Wireless Access (BWA) @ IEEE Globecom 13, 2013. [6] J. Anderson, F. Rusek, and V. Owall, Faster-Than-Nyquist signaling, Proceedings of the IEEE, vol. 101, no. 8, pp. 1817 1830, Aug. 2013. [7] N. Michailow et. al, Generalized Frequency Division Multiplexing for 5th Generation Cellular Networks, IEEE Transactions on Communications, vol. 62, no. 9, pp. 3045 3061, 2014. [8] R. Datta, N. Michailow, M. Lentmaier, and G. Fettweis, GFDM Interference Cancellation for Flexible Cognitive Radio PHY Design, in 2012 IEEE Vehicular Technology Conference (VTC Fall). IEEE, Sep. 2012, pp. 1 5. [9] M. Danneberg, R. Datta, and G. Fettweis, Experimental Testbed for Dynamic Spectrum Access and Sensing of 5G GFDM Waveforms, in Proceedings of the IEEE 80th IEEE Vehicular Technology Conference (VTC Fall 14). [10] M. Matthé, L. L. Mendes, and G. Fettweis, Asynchronous multi-user uplink transmission with generalized frequency division multiplexing, in Proceedings of IEEE International Conference on Communications (ICC) - Workshop on Advanced PHY and MAC Techniques for Super Dense Wireless Networks, London, UK, 2015. [11], GFDM in a Gabor Transform Setting, IEEE Communications Letters, vol. 18, no. 8, pp. 1379 1382, 2014. [12] M. Matthé, L. Mendes, I. Gaspar, N. Michailow, and G. Fettweis, Multi- User Time-Reversal STC-GFDM for 5G Networks, EURASIP Journal on Wireless Communications and Networking, 2015. [13] M. Danneberg, R. Datta, A. Festag, and G. Fettweis, Experimental testbed for 5g cognitive radio access in 4g lte cellular systems, in Sensor Array and Multichannel Signal Processing Workshop (SAM), 2014 IEEE 8th, June 2014, pp. 321 324.