- Research
- Open Access

# A synchronization technique for generalized frequency division multiplexing

- Ivan S Gaspar
^{1}Email author, - Luciano L Mendes
^{2}, - Nicola Michailow
^{1}and - Gerhard Fettweis
^{1}

**2014**:67

https://doi.org/10.1186/1687-6180-2014-67

© Gaspar et al.; licensee Springer. 2014

**Received:**1 December 2013**Accepted:**21 April 2014**Published:**10 May 2014

## Abstract

Generalized frequency division multiplexing (GFDM) is a block filtered multicarrier modulation scheme recently proposed for future wireless communication systems. It generalizes the concept of orthogonal frequency division multiplexing (OFDM), featuring multiple circularly pulse-shaped subsymbols per subcarrier. This paper presents an algorithm for GFDM synchronization and investigates the use of a preamble that consists of two identical parts combined with a windowing process in order to satisfy low out of band radiation requirements. The performance of time and frequency estimation, with and without windowing, is evaluated in terms of the statistical properties of residual offsets and the impact on symbol error rate over frequency-selective channels. A flexible metric that quantifies the penalty of misalignments is derived. The results show that this approach performs practically as state-of-the-art OFDM schemes known in the literature, while it additionally can reduce the sidelobes of the spectrum emission.

## Keywords

- GFDM
- Synchronization
- Windowing
- Metric

## 1 Introduction

Many of today’s digital communication systems use orthogonal frequency division multiplexing (OFDM) [1] as physical layer interface, mainly because of its robustness against frequency-selective channels and easy means of implementation. However, OFDM has some important drawbacks that make it questionable for future wireless systems as 5G networks [2]. In particular, the high out-of-band (OOB) emissions [3] of OFDM signals are an obstacle when using this technology in fragmented spectrum and dynamic spectrum allocation scenarios. Clearly, OFDM is not a straightforward choice for 5G networks, and consequently, new waveforms are being investigated for next-generation standards. However, it is highly convenient if the techniques developed for OFDM can be somehow be applied for new waveforms.

Normally, synchronization techniques for OFDM neglect the effects of the OOB in the preamble design. However, this is an important aspect to be addressed with generalized frequency division multiplexing (GFDM) [4]. Therefore, the main contribution in this paper is the investigation of a solution for the GFDM synchronization that does not increase the OOB emission and has a performance equivalent to the solutions designed for OFDM.

GFDM is a flexible solution to address the requirements imposed by the new scenarios foreseen for the 5G networks [2]. In this scheme, a symbol composed of several subcarriers and multiple subsymbols is used to transmit a data block, where each subcarrier is pulse-shaped with a transmission filter. In order to avoid spending additional samples on ramp up and ramp down of the filter response, the impulse response for each subsymbol is applied through circular convolution. This approach is also called tail biting [4]. Different pulse shapes can be used as prototype filters, which introduce a new degree of freedom to the system. The subcarrier filtering not only reduces OOB emissions but also might introduce intercarrier interference (ICI) and intersymbol interference (ISI). However, receiving techniques such as zero forcing [5] or a matched filter in combination with successive interference cancellation (SIC) [6] can reduce the impact of the self-generated interference and lead to an error rate performance equivalent to OFDM. Implementation approaches based on fast Fourier transform (FFT) algorithm show that it is possible to efficiently construct GFDM transceivers with the technology available today [7].

One of the main challenges in the multicarrier receive chain is symbol time offset (STO) [8] and carrier frequency offset (CFO) estimation [8]. Among the synchronization approaches available for OFDM, data-aided methods allow the use of both autocorrelation and cross-correlation properties, and one-shot synchronization can be achieved in bit pipe and burst communications. A sequence composed of two repeated OFDM symbols was first proposed in [9] in order to explore a strong autocorrelation property for CFO estimation, and shortened data symbols were suggested to address wider CFO estimation range. This idea is addressed by [10] with the construction of a single OFDM symbol constituted with two repeated parts. A coarse STO can be achieved with this proposal as well, but the use of cyclic prefix (CP) and cyclic suffix (CS) creates a plateau effect in the autocorrelation, which introduces ambiguity and reduces the precision of the technique. In [11], a solution using additional integration, i.e., moving average along the CP and CS length, to remove plateau effects is proposed. Nevertheless, the metric has a pyramidal shape, which leads to STO estimation errors in noisy environments. These methods are explored in the algorithm in [12] to obtain a robust CFO estimation, and after correcting the CFO, the integrated autocorrelation method is combined with a cross-correlation, leading to an optimized impulsive time metric. An additional threshold criterion allows the fine estimation of the time offset of the first tap of a multipath channel, based on the statistical properties of the previous samples and a given probability of false alarm [13].

In contrast to the large number of synchronization techniques that are available for OFDM, there is a lack of solutions for GFDM. All mentioned techniques consider a rectangular pulse shape, which produces non-negligible OOB emissions and low decaying spectrum sidelobes. This paper presents an approach for data-aided synchronization scheme for GFDM pinching of block boundary of the preamble with a window function to provide smooth fade-in and fade-out transitions. As additional contribution, this paper also presents a metric that evaluates the effect of residual offset errors in the GFDM symbol. This metric allows quantifying the equivalent signal-to-noise ratio (SNR) penalty caused by residual time and frequency misalignment for different GFDM parameterization, including the OFDM case. In perfect synchronization conditions, this metric reveals the ICI and ISI introduced by a given prototype filter.

The remaining of this paper is organized as follows: Section 2 brings the background about GFDM. Section 3 presents the proposed solutions for time and frequency synchronization of GFDM signals. Section 4 presents the performance results obtained with the non-windowed and windowed schemes, while Section 5 introduces the metric derived to evaluate the impact of time and frequency misalignments in GFDM signals. Finally, Section 6 concludes this paper.

## 2 GFDM background

*K*subcarriers carrying

*M*subsymbols each, which result in a total of

*MK*parallel substreams. The binary data is mapped into complex valued quadrature amplitude modulation (QAM) symbols. Each stream is critically upsampled by a factor

*N*=

*M*

*K*, i.e.,

*N*−1 zeros are appended. These zero-stuffed sequences are filtered with a corresponding transmit filter

*g*[

*n*] denotes the impulse response of a prototype filter with

*N*samples, while

*k*,

*m*, and

*n*are subcarrier, subsymbol, and time sample indices, respectively. This sequence can also be represented as a column vector

**g**

_{k,m}. Notice that the modulo operation in combination with a prototype filter effectively produces a circular convolution of the upsampled data symbols with the filter (tail biting filtering) as depicted in Figure 2. Therefore, the sequence after the filtering process remains of length

*N*samples. Finally, the baseband transmit signal in digital domain is obtained through the summation of all subcarrier and subsymbol signals according to

with *d*_{k,m} being the data symbol transmitted in the *m* th subsymbol of the *k* th subcarrier. A single CP and/or CS can be added to the entire GFDM symbol, leading to the transmit signal $\stackrel{~}{x}\left[\phantom{\rule{0.3em}{0ex}}n\right]$.

**x**=(

*x*[

*n*])

^{T}with

*n*=0,1,…,

*N*−1 can be also expressed as

**A**

^{(e)}=

**w**

**E**

**A**and

is the extension matrix that replicates *L*_{CP} samples for the prefix and *L*_{CS} for suffix, resulting in *L*=*L*_{CP}+*L*_{CS} additional samples. **I**_{
i
} denotes the identity matrices, **0**_{i,j} is an *i*×*j* matrix containing zeros, and **w** is a generic diagonal matrix that contains the *N*+*L* coefficients of a time window function.

where **h** is a realization of a multipath channel, $\stackrel{~}{\mathbf{v}}$ is an AWGN vector, and ∗ denotes convolution. It is assumed that CP and CS are large enough to accommodate the channel delay spread. The received signal after CP and CS removal is denoted by **y**.

and **B** is a receiver matrix.

is able to remove any self-generated interference once **B**_{ZF}**A**=**I**_{
N
}. However, because $\xi =\sum _{i=0}^{N-1}{\left|{\left[\phantom{\rule{0.3em}{0ex}}\mathbf{B}\right]}_{k,i}\right|}^{2}\ge 1$ for any *k*, the ZF receiver can enhance the noise, leading to a performance degradation when compared with OFDM.

The performance gap between the GFDM receivers and the OFDM depends on the prototype filter, which can be designed to minimize this performance loss. The GFDM prototype filter must address also other requirements, such as low OOB emission. Therefore, a trade-off between all requirements must be found during the pulse shape design [14]. A simple example of the impact of the filter design in the introduced MF and ZF receivers will be addressed in Section 5. Details about more advanced GFDM receivers, the linear minimum mean square error (MMSE), and SIC approaches can be found in [5–7].

## 3 Synchronization approach for GFDM

*M*

_{p}=2 subsymbols and

*K*

_{p}subcarriers. In this block, a PN sequence

**c**=(

*c*[ 0],…,

*c*[

*K*

_{p}−1])

^{T}of length

*K*

_{p}is transmitted twice, i.e., the preamble carries the data vector [10]:

**w**=(

*w*[

*n*])

^{T}with ramp-up transition

*w*

_{up}[

*n*] and ramp-down transition

*w*

_{down}[

*n*] is defined. As an example, the corresponding coefficients can be taken from a variation of the Tukey pulse [15]:

*n*=0, 1 ⋯,

*N*+

*L*−1 is the time index,

*W*is the length of the ramp-up and ramp-down transitions, and v(

*n*) is known as the Meyer wavelet auxiliary function [16]:

where ${\mathbf{A}}_{\mathbf{w}}^{\left(\mathrm{e}\right)}$ is the extended modulation matrix from (6), which incorporates the block transitions in **w**.

Note that the preamble sequence presented in [12], which is designed for OFDM, can be reproduced using the GFDM matrix notation ${\mathbf{p}}_{1}={\mathbf{A}}_{\mathbf{1}}^{\left(\mathrm{e}\right)}{\mathbf{d}}_{\mathrm{p}}$, where ${\mathbf{A}}_{\mathbf{1}}^{\left(\mathrm{e}\right)}$ is constructed with a rectangular window, i.e., *w*[*n*]=1for *n*∈{0,*N*+*L*−1}.

The prototype filter *g*[*k*] can also be defined as a rectangular function covering exactly the duration of one subsymbol for both non-windowed and the windowed preamble once the two repeated subsymbols are cyclically connected.

*K*

_{p}=128 and

*L*=

*L*

_{CP}=

*W*=32. In the figure, half of the subcarriers are intentionally disabled to illustrate the difference in OOB emissions between the conventional and the extended approach. While the non-windowed preamble

**p**

_{1}produces non-negligible spectrum sidelobe of −35 dBc, the proposed smoothing window in

**p**

_{w}is able to drastically reduce the OOB several orders of magnitude below what traditional OFDM can achieve.

*r*[

*n*] is collected. Among those, the transmitted preamble needs to be received at least once, in order to be able to estimate time and frequency offsets. An autocorrelation of the sequence

*r*[

*n*] is performed according to [10] as

where *p*_{×}[ *k*] contains the two halves of the original preamble obtained from the operation **p**_{×}=**A** **d**_{p}.

*μ*

_{S}[

*n*],

*μ*

_{M}[

*n*], |

*ρ*

_{ C }[

*n*]|, and

*μ*

_{A}[

*n*] and illustrates the metric evolution in the schemes from [10] to [11] and [12]. Without windowing, the plateau effect in

*μ*

_{S}[

*n*] is eliminated in

*μ*

_{M}[

*n*], and this metric is then multiplied with |

*ρ*

_{ C }[

*n*]| to remove its side peaks. With windowing the final result,

*μ*

_{M}[

*n*]·|

*ρ*

_{ C }[

*n*]| differs only slightly, while the main difference is an alteration in the shape of

*μ*

_{S}[

*n*]. By choosing the window length to be within the CP and CS region, the plateau is suppressed. So, with a proper time offset adjustment, it would be possible to combine

*μ*

_{S}[

*n*] directly with |

*ρ*

_{ C }[

*n*]| and further simplify the algorithm without computing

*μ*

_{M}[

*n*].

Although the impulsive shape of *μ*_{A}[*n*] results in precise STO estimation in single path channels, in time-variant frequency-selective channels, the first multipath peak cannot be found using only the maximum value principle. Fading propagation conditions can affect the initial channel tap gain such that it can be lower than one of the other echoes.

*p*

_{FA}. Based on the central limit theorem, samples that do not correspond to the preamble echoes in the cross-correlation output are interpreted as complex Gaussian random samples, and the threshold is derived from a Rayleigh cumulative density function (CDF), resulting in [13]

where *λ* is an adjustable parameter defined according to the channel impulse response characteristics.

An evaluation on the complexity of the algorithm [12] can be found in [17]. The effect of the refined search for the position of the first multipath will be better evaluated in the next section in terms of the STO and CFO estimation errors.

## 4 Performance analysis of the synchronization algorithm

**Simulation parameters for the preamble**

Parameter | Value |
---|---|

Transmit filter | Rect in time |

Number of subcarriers ( | 128 |

Number of subsymbols ( | 2 |

CP length ( | 32 samples |

CS length ( | 0 samples |

Length of ramp-up/ramp-down ( | 32 samples |

False alarm probability ( | 10 |

Multipath search parameter ( | 32 |

In order to analyze the performance of the proposed windowed preamble, the synchronization technique is evaluated in a time-variant frequency-selective channel. The unknown time offset *θ* and frequency offset *ε* will be estimated considering a channel impulse response with 32 taps. The tap gains vary linearly in decibel scale from 0 to −13 dB. Each tap is multiplied by a Rayleigh distributed random variable with parameter $\sigma =1/\sqrt{2}$.

The Cramér-Rao lower bound is included in Figure 6 and is computed according to [10] to be used as a reference. The gap observed in the CFO estimation, which occurs for both non-windowed and windowed preambles, is due to the coherence loss between the segments of the preamble introduced by the time-variant frequency-selective channel.

The mean squared error of the STO estimation is within tenths of a sample as the SNR is increased for the non-windowed and the windowed preambles. For the particular parameters used in the simulation, it is observed that the mean value of the STO in Figure 7 presents a positive offset behavior that decreases with the increment of the SNR. This is a consequence not only of the multipath delay profile but also of the threshold criterion used for searching the first multipath. The threshold is proportional to the noise level, and if a low false alarm probability is specified, the chance to ignore a severely attenuated first path gets higher.

*μ*

_{M}[

*n*]| is not sharp, this metric can coarsely identify the region where the preamble is present and provides the CFO estimation. Once the CFO is compensated, the cross-correlation |

*ρ*

_{ C }[

*n*]| is obtained. The search for the highest peak is performed with

*μ*

_{A}[

*n*], but it does not represent the first multipath, located in the time sample position 0.

Two regions are explored for a refined search for the position of the first multipath in |*ρ*_{
C
}[ *n*]|: region 1 (${\mathcal{R}}_{1}$) defines the range used to compute the threshold value, and region 2 (${\mathcal{R}}_{2}$) defines the range where peaks higher than the threshold are considered as previous multipaths. The zoomed curve shows the range ${\widehat{n}}_{\mathrm{A}}-\lambda \le n\le {\widehat{n}}_{\mathrm{A}}$. In this example, *h*[ 0]<<*h*[ 1] and the threshold criterion points that *h*[ 1] should be considered as the first multipath, resulting in a positive offset error.

As can be seen in Figure 8, even when the channel delay profile has a weak first tap and the SNR is low (0 dB), the residual estimation error of the position of the primary path is small. Furthermore, in higher SNR conditions, the impact of losing a weak multipath is not as severe as a potential ISI from an adjacent transmitted sequence in a bit pipe communication system, which might happen if a negative residual error is introduced.

## 5 Metric to evaluate synchronization errors

*θ*

_{r}within the CP and CS interval and the residual frequency offset

*ε*

_{r}within one subcarrier bandwidth.

*θ*

_{r}and

*ε*

_{r}, the estimated symbols on the receiver side are

**y**

^{′}is the received signal after synchronization with residual time and frequency offsets, and

**G**is the overall system matrix. The matrix that represents the overall residual time and frequency offsets is given by

**F**is the Fourier matrix,

is the matrix representing the time shifts.

**G**=

**I**

_{ N }, the recovered symbols equal the transmitted symbols. Hence, the mean square error between

**G**and the identity matrix can be seen as a measure of the degradation introduced by the self-generated interference in a non-orthogonal system and the time and frequency offsets. In this sense, a metric to evaluate the interference caused by residual synchronization errors can be defined by

**Θ**=

**I**

_{ N }, the metric can also be used to evaluate the prototype filter

*g*[

*n*] and the performance of the system,

**B**

**A**. For instance, Figure 10 exhibits the influence of the MF receiver with raised cosine (RC) and root-raised cosine (RRC) [14] as a function of the roll-off parameter

*α*, assuming perfect synchronization.

From Figure 10, it is possible to conclude that small roll-off values leads to smaller interference, while RC filter outperforms RRC for any value of *α*. This is an interesting counterintuitive result, because the interference introduced by the MF with RC is composed by ISI and ICI, while the RRC is a Nyquist pulse and thus creates only ICI [14].

When the roll-off is equal to zero, the self-interference is absent, both RRC and RC became a rect filter in frequency domain. This is a corner case where GFDM implements the orthogonal concept of the localized single-carrier frequency-division multiple access (SC-FDMA) scheme [18].

*ξ*to compute the impact of both the prototype filter and synchronization errors in a given SNR, as

where *E* is the average energy of the signal, and *N*_{0} is the noise power spectral density.

*M*=1 and

*g*[

*k*] is a rectangular prototype filter, is presented in order to be compared with the lower bound [9]

Figure 11 shows that the metric (32) is consistent with the lower bound proposed in [9] for the OFDM case, although GFDM has further SNR decrement because of the noise enhancement or self-interference. Furthermore, if the impact of self-generated interference on the metric value is known, then (32) can be used to evaluate the impact of synchronization errors in the system performance. Notice that the self-interference degradation can be easily obtained from (32) by assuming perfect synchronization, as shown in Figure 10. When using ZF, the noise enhancement *ξ* is also smaller employing the RC than RRC, and it can be even neglected if *α*→0.

**Configuration of the GFDM data block**

Parameter | Value |
---|---|

Mapping | 16-QAM |

Transmit filter | RC |

Roll-off ( | 0.1 |

Number of subcarriers ( | 128 |

Number of subsymbols ( | 9 |

CP length ( | 32 samples |

CS length ( | 16 samples |

Length of ramp-up/ramp-down ( | 16 samples |

The main difference between the SER obtained assuming the synchronization approaches and the curve assuming perfect synchronization is that the high variance of the CFO in low SNR regime leads to a performance gap when the synchronization techniques are considered. However, all curves tend to match each other for high SNR, since in this case, the CFO variance became neglectable, as shown in Figure 6.

## 6 Conclusions

In this paper, time windowing has been proposed in combination with a synchronization preamble in order to preserve the advantageous spectral properties in a GFDM transmission. A window has been presented, which is able to provide a considerable reduction in the spectrum sidelobes, several orders of magnitude below that can be achieved with a standard preamble designed for OFDM.

The results have further shown that the windowing can contour the plateau effect in the autocorrelatiios foreseen for the next-generation wireless networks, where standard OFDM approaches are not suitable.

## Declarations

## Authors’ Affiliations

## References

- Savaux V, Louët Y, Djoko-Kouam M, Skrzypczak A: Application of a joint and iterative MMSE-based estimation of SNR and frequency-selective channel for OFDM systems.
*EURASIP J. Adv. Signal Process*2013, 2013(1):128. 10.1186/1687-6180-2013-128View ArticleGoogle Scholar - Wunder G, Jung P, Kasparick M, Wild T, Schaich F, Chen Y, Brink S, Gaspar I, Michailow N, Festag A, Mendes L, Cassiau N, Ktenas D, Dryjanski M, Pietrzyk S, Eged B, Vago P, Wiedmann F: 5GNOW: non-orthogonal, asynchronous waveforms for future mobile applications.
*IEEE Commun. Mag*2014, 52: 97-105.View ArticleGoogle Scholar - Beek JVD, Berggren F: Out-of-band power suppression in OFDM.
*IEEE Comm. Lett*2008, 12(9):609-611.View ArticleGoogle Scholar - Fettweis G, Krondorf M, Bittner S:
*GFDM - generalized frequency division multiplexing. Paper presented at the IEEE 69th vehicular technology conference, Hilton Diagonal Mar.*Barcelona, Spain; 26–29 April 2009:1-4.Google Scholar - Michailow N, Krone S, Lentmaier M, Fettweis G:
*Bit error rate performance of generalized frequency division multiplexing. Paper presented at the 76th, IEEE vehicular technology conference (VTC Fall’12).*Québec, Canada; 3:1-5.Google Scholar - Datta R, Michailow N, Lentmaier M, Fettweis G: GFDM interference cancellation for flexible cognitive radio PHY design.
*VTC*2012, 1: 1-5.Google Scholar - Gaspar I, Michailow N, Navarro A, Ohlmer E, Krone S, Fettweis G:
*Low complexity GFDM receiver based on sparse frequency domain processing. Paper presented at the, IEEE 77th vehicular technology conference (VTC Spring).*Dresden, Germany; 2–5 June 2013:1-6.Google Scholar - Bahai ARS, Saltzberg BR, Ergen M: Synchronization. In
*Multi-Carrier Digital Communications: Theory and Applications of OFDM*. Springer, New York; 2004:83-101.Google Scholar - PH Moose: A technique for orthogonal frequency division multiplexing frequency offset correction.
*IEEE Trans. Comm*1994, 42(10):2908-2914. 10.1109/26.328961View ArticleGoogle Scholar - Schmidl T, Cox D: Robust frequency and timing synchronization for OFDM.
*IEEE Trans. Comm*1997, 45(12):1613-1621. 10.1109/26.650240View ArticleGoogle Scholar - Minn H, Zeng M, Bhargava V: On timing offset estimation for OFDM systems.
*IEEE Commun. Lett*2000, 4(7):242-244.View ArticleGoogle Scholar - Awoseyila A, Kasparis C, Evans B: Improved preamble-aided timing estimation for OFDM systems.
*IEEE Commun. Lett*2008, 12(11):825-827.View ArticleGoogle Scholar - Kasparis C, Evans BG:
*A cross-correlation approach for improved timing estimation in, OFDM broadcasting systems. Paper presented at the 24th AIAA international communications satellite systems conferences (ICSSC),.*San Diego, CA, USA; 11–14 June 2006:1039-1048.Google Scholar - Matthé M, Michailow N, Gaspar I, Fettweis G:
*Influence of pulse shaping on bit error rate performance and out of band radiation of generalized frequency division multiplexing. Paper presented at the ICC’14 - workshop on 5G technologies (ICC’14 WS - 5G),.*Sydney, Australia; 10.Google Scholar - Bloomfield P:
*Fourier Analysis of Time Series: An Introduction*. Wiley-Interscience, New York; 2000.View ArticleGoogle Scholar - Meyer Y:
*Ondelettes et opérateurs: Ondelettes*. Actualités mathématiques, Hermann; 1990.Google Scholar - Abdzadeh-Ziabari H, Shayesteh MG: Robust timing and frequency synchronization for ofdm systems.
*IEEE Trans. Veh. Tech*2011, 60(8):3646-3656.View ArticleGoogle Scholar - Myung HG, Lim J, Goodman D: Single carrier FDMA for uplink wireless transmission.
*IEEE Veh. Tech. Mag*2006, 1(3):30-38.View ArticleGoogle Scholar

## Copyright

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.