GFDM Frame Design For 5 G Application Scenarios

The services foreseen for 5G networks will demand a vast number of challenging requirements to be fulfilled by the physical layer. These services can be grouped into application scenarios, each one with a key requirement to be addressed by the 5G network. A flexible waveform associated with a appropriate data frame is an essential feature in order to guarantee the support of contrasting requirements from different application scenarios such as Enhanced Mobile Broadband, Internet of Things, Tactile Internet and Internet Access for Remote Areas. In this paper, we propose a flexible data frame based on Generalized Frequency Division Multiplexing (GFDM) that can be tailored to address the specific key requirements of the different 5G scenarios. The paper also presents the physical layer parametrization that can be used for each application.


I. INTRODUCTION
C ELLULAR terminals have evolved from devices provid- ing voice connectivity to source and sink for multimedia, sensor, positioning and other varied digital content.Nowadays, several new application scenario are being considered for fifth generation of mobile network (5G) and each one has a key requirement to be addressed.Enhanced Mobile Broadband (eMBB) is the natural evolution of fourth generation of mobile network (4G), where the aim is to provide data rates up to 10 Gbps [1].Ultra Reliable Low Latency (URLL) applications, also known as Tactile Internet [2], requires latency below 1 ms and robustness in order to avoid packages retransmissions.The Internet of Things (IoT) and Machine-Type Communication (MTC) will demand connectivity of a massive number of energy-limited devices per cell [3], which cannot afford cumbersome synchronization process with the network.Internet Access for Remote Areas (IARA) is the last frontier for universal connectivity and 5G must present a operation mode capable of supplying Internet connect in remote/rural areas, where the cell coverage must be bigger than 50 km [4] and, therefore, over a doubly-dispersive channel [5].
It is clear that these contrasting requirements do not need to be simultaneously fulfilled.5G networks are expected to have different operation modes, each one designed for one specific application scenario.The base station (BS) computational resources, as well the time and frequency grid need to be dynamically allocated for the different application scenarios, depending on the users' demands.A new concept, defined as network slicing [6], allows for sharing the communication and processing resources of a 5G BS among the different 5G services.Each operation mode could have a specific physical layer (PHY).However a unified data frame, based on a flexible waveform able to address the different 5G requirements would simplify the network management and deployment of future services.Therefore, the approach based on a single waveform is preferred over the approach based on a multitude of specific PHYs.
Several new waveforms have been recently proposed for 5G networks [7].Filterbank Multicarrier (FBMC) [8] relies on linear filtering per subcarrier and Offset Quadrature Amplitude Modulation (OQAM) to mitigate the intercarrier interference (ICI), achieving high spectrum efficiency and very low out-ofband emission (OOBE).However, the long filter tails hinder the usage of this waveforms in situations where a short burst of data needs to be transmitted, such as MTC and IoT applications.Universal Filtered Multi-Carrier (UFMC) [9], also known as Universal Filtered Orthogonal Frequency Multiplexing (UF-OFDM), is an Orthogonal Frequency Division Multiplexing (OFDM) [10] based waveform, where a short length filter is used to filter a set of subcarriers, which remain orthogonal to each other within the filter's subband.UFMC does not use cyclic prefix (CP) to combat intersymbol interference (ISI) and zero-padding is employed to accommodate the filter spreading.Therefore, UFMC is more sensitive to time misalignments, resulting in performance loss even when small time synchronization errors occur [9].Filtered Orthogonal Frequency Division Multiplexing (F-OFDM) [11] is another OFDM-based waveform that also employs subband filtering.But, unlike UFMC, F-OFDM uses CP to avoid ISI in multipath channels.F-OFDM can achieve low OOBE and the CP allow for small time synchronization errors without performance penalty.However, like OFDM, F-OFDM employs one CP per symbol, reducing the spectrum efficiency specially when short symbols are required, i.e., in URLL applications.Generalized Frequency Division Multiplexing (GFDM) [12] is another 5G candidate waveform that, like FBMC, employs subcarrierbased filtering to reduce OOBE.However, GFDM uses circular filtering in time and frequency domains, meaning that the filter spreading is confined with the GFDM symbol.In this case, K subcarrier, each one composed by M subsymbols, are used to transmit N = K M quadrature amplitude modulation (QAM) data symbols.CP and cyclic suffix (CS) can be used to protect the symbols against the multipath channels.
Among these candidates, GFDM has been shown to represent a trade-off between complexity and performance, as-suming the different requirements imposed on 5G networks [13].GFDM symbol can be shrunken to reduce latency, while keep the efficient usage of the CP, and it can be precoded with Walsh-Hadamard Transform (WHT) to increase robustness for URLL application [14].Time-windowing and blank subsymbols [15] can be used to reduce even further the OOBE, trigging dynamic and fragmented spectrum allocation [16] required for IARA and eMBB applications.The usage of CP and CS protects the GFDM symbol of small timing synchronization errors.Because of the high flexibility and the acceptable performance for all foreseen application scenario, we have selected GFDM as the waveform to be employed in the 5G frame design.
The aim of this paper is to present a flexible PHY frame, based on the GFDM features, that can be used to properly address the requirements of the different 5G scenarios.The proposed PHY frame can be adjusted to achieve high spectrum efficiency, low latency, synchronization robustness and high protection against multipath channels.The paper also brings a set of PHY parameters that can be employed to support each application scenario, showing that an unified PHY frame based on GFDM can be used to fulfill the 5G demands.
This paper is organized as follows: section II brings the application scenarios and the corresponding requirements, while section III presents the GFDM principles.Section IV describes the frame design methodology and section VI proposes the parametrization of the PHY for each application scenario.Finally, section VII concludes the paper.

II. SCENARIOS FOR 5G
The main characteristic of 5G network is that it must support services that differ greatly in their communication requirements and performance targets.These services can be organized in application scenarios sharing similar requirements in terms of throughput, latency and reliability [1] [17].The main 5G application scenarios are briefly described next.

A. Enhanced Mobile Broadband (eMBB)
This scenario can be seen as an evolution of the current 4G networks, since its main objective is providing higher throughput of up to 10 Gbps.End-to-end latency should not exceed 10 ms for multimedia traffic.Cell radius of up to one kilometer can be assumed in urban areas.The main techniques that are being considered to support these requirements are millimeter wave combined with massive multiple-input multiple-output (MIMO) [18] [19], cooperative multi-point (CoMP) [20] and high spectrum efficient waveform with low OOBE [12].

B. Ultra Reliable Low Latency (URLL)
The mission-critical URLL aims for high reliable low latency services, where the low end-to-end latency must be less than 1 ms.This requirement means that the overall frame size must be limited, implying in a restricted forward error control (FEC) length [21].This is a very challenging situation, since a robust FEC must guarantee low bit error rate (BER) and Automatic repeat request (ARQ) [22] cannot be used due to the latency requirement.

C. Internet of Things (IoT) / Machine-Type Communication (MTC)
IoT/MTC will play a key role in the definition of 5G networks.This scenario is being pointed as the main opportunity for new revenues for telecommunication operator.The number of devices connected to the network (over 50,000 devices per cell) [1] and long battery life time (above 10 years) are the key requirements of this scenario.Low energy consumption, loose synchronization procedure and random channel access are some features that the PHY must provide for this application scenario.The peak data rate considered is between 1 kbps to 10 Mbps.Latency requirements vary from 1 to 100 ms in order to cover a broad range of applications, from the demanding industrial automation to smart metering.

D. Internet Access for Remote Areas (IARA)
IARA basically demands large cellular coverage in order to increase the number of subscribers per cell in rural and low populated areas.Opportunistic dynamic and fragmented spectrum allocation is a key feature in this scenario.The PHY must be based on a waveform that can efficiently exploit the CP in order to achieve high spectrum efficiency under channels with long delay profile.Also, the peak data rate and end-to-end latency that it can provide should be comparable to Long-Term Evolution (LTE), i.e. 100 Mbps and latency of 50-100 ms.
Table I summarizes the requirements of foreseen 5G application scenarios.

III. GFDM MODEL
Assume that a GFDM block occupies K subcarriers, each one carrying M data symbols, which are temporally equally spaced and circularly filtered by a prototype filter g[n] with N = K M samples.The GFDM block is given by where d k,m is a QAM symbol, a b stands for the a modulo b operator and K on and M on are the sets of active subcarrier and subsymbols, respectively.As illustrated in Fig. 1, it is possible to arbitrarily insert the QAM symbols in the GFDM timefrequency grid.Hence, a GFDM symbol carries N on = M on K on data symbols, where M on and K on are the cardinalities of K on and M on , respectively.Analogous to OFDM, GFDM also adopts CP to mitigate multipath channels, but a single CP is used to protect Fig. 1.Time-frequency grid of GFDM and terminology.
subsymbols, significantly increasing the spectrum efficiency when M on is large, e.g.M on ≥ 5.
Another interesting GFDM feature is that this waveform can cover the 4G waveforms, i.e., OFDM and DFT spread OFDM (DFT-s-OFDM), as corner cases.OFDM is obtained by making M = 1 and using a square pulse as prototype filter, while DFT-s-OFDM is obtained when K = 1 and using a Dirichlet pulse as prototype filter.Therefore, GFDM can also be employed in hybrid 4G-5G system during the standard transition, smoothing the transition to a new standard, while still offering support for legacy users.
GFDM offers additional degrees of freedom for enhancing the waveform performance.Peak to average power ratio (PAPR) of DFT-s-OFDM can be reduced by increasing the roll-off factor [23] and the OOBE can be reduced by leaving blank subsymbols in the head and tail of the GFDM block [12].Time-domain windowing can further reduce the OOBE [15].Using circular rather than linear filtering, GFDM produces symbols that are well localized in time, which makes GFDM robust against time misalignment.Hence, the flexibility offered by GFDM makes it an attractive solution to achieve a unified air interface for diverse 5G scenarios.

IV. GFDM-BASED FRAME: PROPOSED DATA STRUCTURE
The PHY structure for the 5G PHY must allow a flexible resource allocation into the M on subsymbols of the the K on subcarriers, considering all required overhead, such as coding rate, synchronization preambles, CP and CS. Figure 2 depicts the frame structure proposed in this paper and the terminology used to identify each element of the frame.
The code word length is an important parameter with a trade-off between correction performance and latency.The final code word is mapped into Consider a bit sequence with length k L representing the uncoded PHY payload, which consists of user data and upper layer headers.A channel coding of rate R C = k L /n L is employed to form a code-word with length of n L bits.The bits of the code-word are mapped into a M C -QAM constellation, where M C is the modulation order.This QAM sequence is defined as FEC Block.Fig. 2.a describes this data flow.It is assumed that n L is fixed for a given scenario.
Let's N QAM = n L /log 2 (M C ) be the number of QAM symbols within a FEC Block.Clearly, N QAM is not necessarily equal to the number of the active data symbols N on within one GFDM symbol.Therefore, N FEC FEC blocks, containing N FEC N QAM QAM symbols, are mapped into N G GFDM symbols (or GFDM resource blocks), which has the capacity to carry N G N on QAM symbols, as shown in Fig. 2.b.Therefore, must hold.
The frame is generated based on the smallest N FEC that satisfies (2).This can be evaluated as where LCM(a, b) returns the Least Common Multiple of a and b.Now, considering that preamble-based synchronization and channel estimation are employed, a frame is generated by time-multiplexing N G modulated GFDM symbols (N FEC FEC blocks) with N P preambles.The waveform employed as preamble is known by the receiver and it has the same spectral and time properties of the GFDM signal.The preamble spectral content spreads through the desired bandwidth and the time interval between two successive preambles must be smaller than the channel coherence time.Clearly, increasing the number of preambles per frame eases the synchronization process and improve channel tracking, but at the cost of throughput reduction.The frame duration is given by where T P is the preamble duration and T G is the GFDM symbol duration, which is expressed as with S GS being the number of samples per GFDM symbol and F S being the sampling frequency.S RB is the number of MIMO is an essential technique for 5G systems.Time-Reverse Space Time Coding (TR-STC) [24] can be applied on groups of GFDM symbols, in order to achieve transmit diversity.The data frame proposed in this paper allows for adjusting the number of preambles that are transmitted by each antenna.On the proposed scheme, all transmit antennas sends preambles simultaneously, assuming that the receivers have two or more receive antennas.TR-STC is not applied on the preambles and the channel state information can be estimated on all receiving antennas using minimum mean square error (MMSE) or zero-forcing (ZF) [12], depending on the complexity that the receiver can afford.
Finally, an integer number of frames are concatenated, resulting in a super-frame as depicted in Fig. 2.d, which has the exact duration of 1 second.Assuming the base station have access to the global positioning system (GPS) system, the super-frame may be locked to the 1 pulse per second (1PPS) signal.When the receiving device also has access to the GPS signal, time and frequency synchronization is easily acquired with a 1PPS-aided scheme.Likewise, when the receiver is synchronized with the base station, a fast GPS location is acquired with a GFDM-aided scheme.Usually, tens of seconds are spent to discipline a local oscillator based on the received 1PPS signal, in order to acquire an accurate location.The energy and time spent on synchronization can be safe if a local oscillator derived from the GFDM symbol is already locked to the GPS system.
In order to guarantee the super-frame fits exactly in 1 second, GFDM symbol and preamble duration time must be accurately adjusted by the varying number of samples in CS and/or CP.Therefore, the integer number of frames in a superframe is given by where T S is the super-frame duration, which is fixed to 1 second for convenience.
V. PERFORMANCE CRITERIA FOR 5G The parametrization of the GFDM frame for each 5G application scenario depends on the key requirement that needs to be addressed.Following we describe the relationship between the requirements and the system parameters.
1) System bandwidth: The bandwidth of the resultant GFDM signal is given by 2) PHY Gross bit rate: The total bit rate transmitted by GFDM symbols is given by 3) Spectral efficiency: The spectral efficiency of the system is given by ε = R U /B W . Based on Shannon's theorem for channel capacity and assuming a coding scheme that leads to the optimum performance, the minimum signal-to-noise ratio (SNR) required for a given spectral efficiency under additive white Gaussian noise (AWGN) channel [25] is 4) PHY Latency: The system latency can be estimated based on the time transmission interval (TTI), which is the time interval required to the PHY to transmit one code-word.The average TTI can is approximately give by The one-way transit time of a FEC block can be roughly approximated by 3.5T TTI [26].Therefore, the round trip time (RTT) [27] is T RTT ≈ 7T TTI .5) Mobile speed: Considering channel state information is estimated on every preamble, the maximum supported mobile speed is determined by the time interval between preambles, which is given by The minimum channel coherence time is T C = ∆T P /0.7, assuming a 70% confidence factor.The maximum Doppler frequency is f D max ≈ 0.423/T C [28] and the maximum supported speed is , where λ c is the carrier wavelength [28].
6) Battery lifetime: In order to establish an approximate estimative of the impact from the frame design into the power consumption, a simplified consumption model based on [29] and [30] was applied on a mobile autonomous reporting (MAR) IoT device [31].We assume periodic reports transmission without ARQ.In the adopted model, a MAR device stays on sleep mode most of the time and bursts data following the steps described next: i) Warm-up to wake up from sleep; ii) Time and frequency synchronization; iii) Transmission of one FEC block (TTI); iv) Turning off the RF circuitry; and v) sleep considering 1 transmission per hour.The power consumption analysis mentioned in this paper considers a 3.3 V battery with capacity of 6, 500 Joules [30].

VI. GFDM-BASED FRAME: MEETING THE REQUIREMENTS OF EACH SCENARIO
Different 5G operation modes are necessary in order to address the key requirements of the application scenarios.In Table II, we propose four different parametrization of the GFDM frame presented in this paper.Each set of parameters aims for attending a specific 5G scenario.The preamble used in this propose consists on a GFDM symbol with M = M on = 2.The preamble load is formed by a sequence of K on QAM symbols, which are repeated in each subsymbol.Tis preamble content creates a time-redundancy that can be exploited by the synchronization algorithm on the receiver side [32].
Tables III and IV summarize the frame parameters for each scenario and the achieved performance, respectively.The parameter R k in Table IV represents the multipath robustness of the GFDM symbol, obtained as where c is the speed of light in vacuum.A 10 GHz carrier frequency has been considered to estimate V m .R k were defined considering that the multipath robustness is equivalent to 15 to 18% of the cell radius [33].The preambles length were defined using this reasoning.Table V shows the performance parameters considering the maximum number of users or devices allocated simultaneously for the set of parameters given by Table II.It was considerate one carrier as guard between each user, resulting on the total number of carriers per GFDM Symbol K on (Max.users) in Table II, and the total number of active data symbols transported by each GFDM Symbol N on (Max.users) in Table III.In Table Table V, U s represents the maximum number of users allocated per frame and R u represents the associated data rate per user.U r give the maximum rate allocation of users per second.Finally, the parameter U b gives the maximum rate allocation of users per second per Hertz, obtained by U r /B W .
Next subsections summarize the achieved results for each application scenario.

A. eMBB parameters and performance analysis
The number of subcarriers, number of subsymbols, and sampling frequency were selected to be compatible with LTE [34].The considered channel coding is an LDPC (64800, 54000) with R C = 0.83.The 256-QAM is used to increase spectral The suggested configuration provides the highest bit rate and spectral efficiency among the considered scenarios.In order to comply with the 10 Gbps specification from Table I, it is necessary to consider multiplexing MIMO and wider channel bandwidth.Considering an available bandwidth of 200 MHz and the spectral efficiency of 6 bps/Hz, it is necessary to employ an 9 × 9 MIMO in order to achieve the expected peak data rate for this scenario.Massive MIMO scheme with considerably higher arrays has already been demonstrated [35].

B. URLL parameters and performance analysis
The key requirement in this scenario is latency.The selected code-word length is small, with 480 bits, in order to decrease the overall latency.Considering the receiver must wait for the complete GFDM symbol before demodulation takes place, the symbol duration T G has to be short.This is achieved by adopting a small number of subcarriers and subsymbols for a given sampling frequency.Furthermore, the number of active symbols N on is optimized to carry one FEC frame composed of N QAM QAM symbols, thus minimizing the TTI.The suggested configuration provides the lowest latency in terms of RTT with T RTT = 0.42 ms, being the smallest among the selected scenarios.The resulting TTI is T TTI = 0.06 ms, being 16.7 times smaller than the LTE TTI.

C. IoT/MTC parameters and performance analysis
Considering IoT and MTC devices must present low power consumption and low cost, the low sampling frequency of 0.96 MHz was adopted, being a sub-multiple of 30.72 MHz.This procedure reduces the bandwidth while maintaining backward compatibility with LTE master clock.Note that GFDM was parameterized with K on = 1, leading to a single carrier system with low PAPR.The code-word length is 480 bits, complying with short messages typically sent by IoT devices.
The battery life-time estimated for a IoT device operating in the proposed system is 13.45 years, calculated in according with the power consumption model presented in Section V, surpassing the required 10 years life time.
Considering the minimum resource allocation for one device is 1 subcarrier per frame,up to 8 devices can share one frame.The maximum number of devices that cab be allocated per second is evaluated as 8N F , resulting in up to 12288 different devices per second, given by U r .Therefore, this scenario presents the best U b among the considered scenarios.It complies with a big number of devices per cell required on this scenario.

D. IARA parameters and performance analysis
The main specification of the IARA scenario is the extended coverage radius, for which a long CP is required to deal with multipath propagation.The OOBE is another critical requirement, since opportunistic spectrum allocation may be required to reduce the spectrum costs.Time windowing, combined with GFDM, is used here to achieve low OOBE.We assume that the maximum propagation distance between different paths without generating inter-symbol interference is 20 km.The obtained cell radius complies with the required wide coverage area.

VII. CONCLUSIONS
The design of the next generation of cellular networks needs to provide support for a new set of services, organized in application scenarios, that demand very different communication requirements.The proposed frame structure exploits the GFDM flexibility to get a single transmission framework adapted to different services.A set of frame configurations were proposed to comply with each scenario and its analysis shows that the key requirements of the next generation of cellular networks can be effectively achieved by a flexible frame structure and waveform.
Juliano Silveira Ferreira received the B.Sc. and M.Sc. in electrical engineering from Inatel, Brazil, in 2002 and 2009, respectively.Since 2004 he works as a researcher and system specialist at Inatel, Brazil, were has been worked on several research projects funded by FAPEMIG and FINEP, besides projects in partnership with companies.His areas of interest include signal processing and communication systems engineering.Currently he is working on multicarrier modulations for 5G networks and future mobile communications systems.

TABLE I 5G
REQUIREMENTS.
Fig. 2. GFDM-based Frame Design a) FEC Block composition.b)FECBlockencapsulation on GFDM Resource Blocks; c) GFDM Symbol sample length; d) Frame and Super-Frame composition and time duration.samplesperGFDMresource block, S CP e is the number of CP samples effectively used to combat the multipath channel, S CP W and CS W are the number of samples of the CP and CS, respectively, that are used for time-windowing.Fig.2.c illustrates the composition of a GFDM symbol.Similarly T P can be calculated from its corresponding CP and CS parameters.

TABLE II GFDM
-FRAME PARAMETRIZATION FOR THE 5G SCENARIOS.

TABLE III GFDM
FRAME SPECIFICATION FOR EACH SCENARIO.

TABLE IV SYSTEM
PERFORMANCE FOR EACH SCENARIO.

TABLE V SYSTEM
PERFORMANCE FOR MAXIMUM NUMBER OF USERS.