You are on page 1of 5

Adaptive modulation and coding for BIC-UFMC

and BIC-OFDM systems taking CFO into account.


E. Debels, P. Del Fiorentino, C. Vitiello, J. Van Hecke, F. Giannetti, M. Luise, V. Lottici and M. Moeneclaey
TELIN Dept., Universiteit Gent, Sint-Pietersnieuwstraat 41, B-9000 Gent, Belgium
Information Engineering Dept., University of Pisa, I-56122 Pisa, Italy.
E-mail: Erica.Debels@UGent.be, p.delfiorentio@ing.unipi.it, carmine.vitiello@for.unipi.it, filippo.giannetti@iet.unipi.it
marco.luise@unipi.it, vincenzo.lottici@iet.unipi.it, Jeroen.Vanhecke@UGent.be, Marc.Moeneclaey@UGent.be

Abstract—The performance of multicarrier systems can be However, despite the better performance of UFMC com-
considerably degraded because of inter-carrier interference (ICI) pared to OFDM in the presence of a CFO, the performance in
caused by a carrier frequency offset (CFO) between the trans- terms of GP, with the GP defined as the number of correctly re-
mitter and the receiver. A recently developed multicarrier mod-
ulation technique, referred to as uniform filtered multicarrier ceived information bits per unit of time, still decreases rapidly
(UFMC), improves the robustness against CFO, thereby relaxing with increasing CFO, even when the subband filter length has
the synchronization requirements, consequently providing energy been optimized in the presence of a CFO [5], [9], [14]. The
efficient and low latency transmissions. However, the goodput main cause for this reduced GP is the adaptive modulation and
(GP) of the system still rapidly drops with increasing CFO coding (AMC) algorithm ignoring the interference caused by a
when using the classical adaptive modulation and coding (AMC)
schemes based on the SNRs of the subcarriers, because they CFO, yielding a higher packet error rate (PER) than anticipated
ignore the presence of the CFO-induced ICI. To tackle this because of selecting too large constellation sizes and code
problem, this contribution performs the AMC, also taking into rates. In this contribution we modify the AMC algorithm
account the ICI caused by the CFO, and thereby achieving a by taking into account the interference caused by the CFO;
significant increase of the GP. more specifically, the signal to interference plus noise ratio
Index Terms—UFMC, BIC, Adaptive modulation and coding,
Goodput, CFO (SINR) instead of the signal to noise ratio (SNR) of each
individual subcarrier will be used in the estimated GP metric.
This method will reduce the PER by selecting the constellation
I. I NTRODUCTION
sizes and code rate in accordance with not only the amount
The fast growing network of devices connected to the inter- of noise but also interference, which results in a significant
net leads to the need of a fifth generation (5G) communication improvement of the GP performance.
network which is able to deal with diverse traffic types, such The contribution organizes as follows: Section II outlines
as short sporadic packets and normal broadband traffic [1], [2], the system and channel model of the packed-oriented systems
[4], [5]. In this context, the fourth generation (4G) waveform, bit interleaved coded (BIC) -UFMC and BIC-OFDM. Section
orthogonal frequency division multiplexing (OFDM), may not III describes the effect of the CFO and calculates the SINR
be the best choice. Due to its high spectral sidelobes, a when a CFO is present. Section IV shows the GP performance
strict synchronization is necessary to maximize the OFDM when including the SINR in the AMC algorithm, while section
system performance. This strict synchronization requirement V draws the conclusions of this paper.
will be a huge drawback for certain new types of wireless Throughout the paper the following notations are used:
communication such as the Internet of Things (IoT), where lowercase and uppercase bold symbols refer to time-domain
low cost and low latency requirements will not allow for and frequency-domain vectors, respectively; [·]T denotes the
complex synchronization algorithms [6]. Because of its high transpose operator, IN is the N xN identity matrix, |D| is the
sensitivity to a poorly compensated carrier frequency offset cardinality of the set D, x is the lower integer of x, and
(CFO), OFDM could become unsuitable for future generation x ∼ Nc (0, R) indicates that x is a complex-valued circular
networks. A promising solution proposed to overcome this symmetric Gaussian random vector with zero mean and auto-
problem is the universal filtered multicarrier (UFMC) system, covariance matrix R.
which is a trade-off between filter-bank based multicarrier
(FBMC), where every subcarrier is individually filtered, and II. S YSTEM MODEL
OFDM, where filtering is not applied. UFMC is a novel modu- We consider the transmission of packets, with a length of
lation technique where subcarriers are grouped into subbands, Nu bits. These Nu bits consist of Np bits of payload and
in which IFFT and filtering are performed [7]. It has the NCRC cyclic redundancy check (CRC) bits. The bit interleaved
advantage over OFDM of reducing the side lobe level and coded (BIC) modulation block encodes the Nu bits using a
the ICI in case of a CFO, and has the advantage over FBMC code of rate r, selected from the set Dr = {r0 , r1 , ..., r|Dr |−1 },
of being able to use shorter filters, thereby decreasing the and interleaves the coded bits. The interleaved coded bits are
complexity [8]. allocated to N subcarriers according to a bitloading vector

978-1-5090-4361-3/16/$31.00 2016
c IEEE
m = [m, ..., mN −1 ]T with mn ∈ Dm = {0, ..., mmax }, where
mn denotes the number of bits allocated to the nth subcarrier. ࢄ૙ ܰிி் െ ‫࢞ ܶܨܨܫ‬૙ ˆ‹Ž–‡” ࢠ૙ ࢅ

… …
0
The bits on the nth subcarrier are Gray mapped onto complex 0 P/S ࢗ૙ ࢝
symbols from the unit energy 2mn -QAM constellation. The 0

……

0
ܰிி் െ ‫࢞ ܶܨܨܫ‬૚ ˆ‹Ž–‡” ࢠ૚ …Šƒ‡Ž ࢟
code rate r and the bitloading (BL) vector m are selected ࢄ૚ S/P

……
ʹܰிி் -FFT

… …
0 P/S ࢗ૚ ࢎ
0
according to the AMC algorithm [10].


In the case of UFMC, the resulting symbol vector 0
ܰிி் െ ‫࡮ࢠ ”‡–Ž‹ˆ ࡮࢞ ܶܨܨܫ‬


0
X=[X(0), ...., X(N − 1)]T is split over B subbands, each ࢄ࡮ P/S ࢗ࡮


composed of D = N/B subcarriers. The symbols assigned
to the ith subband (with i ∈ {0, 1, ..., B − 1}) are given by
Xi = [X(iD), ..., X((i + 1)D − 1)]T . To each Xi , an IFFT
Fig. 1. UFMC architecture
of size NFFT is applied, with the symbols from the other
subbands set to zero, yielding B time-domain signals xi , i.e.,

NFFT
−1 ࢝
1 −j2π Nmn

xi (m) = √ Xi (n)e FFT (1)
NFFT n=0
ܰிி் -IFFT ࢞

…Šƒ‡Ž CP removal

……

……
S/P

……
CP insert.
for m ∈ {0, 1, ..., NFFT − 1}; note that Xi (n) = 0 for P/S ࢎ ܰிி் -FFT
∀n ∈/ {iD, iD + 1, ..., iD + D − 1}. The signals xi are then
filtered by a frequency-shifted Dolph-Chebyshev filter with
finite impulse response (FIR), qi , of length L, realizing 60 dB
sidelobe suppression; the frequency shift for the ith subband Fig. 2. OFDM architecture
equals ωi = D−12 + iD subcarrier spacings. The resulting B
time-domain signals are summed and sent over the channel.
The duration of one UFMC symbol is NFFT + L − 1 samples. of UFMC. However, in case of zero CFO, this interference can
In the case of OFDM, a NFFT -point IFFT is applied to the ˜
be neglected [9], so we set I(n) = 0 in the sequel.
symbol vector X, and the resulting time-domain signal is sent For OFDM, the receiver keeps the first N samples of the
over the channel after adding a cyclic prefix (CP) of LCP NFFT -point FFT, which are given by
samples.
The multicarrier symbols are sent over a frequency-selective
fading channel, characterized by its impulse response h with Y (n) = H(n)X(n) + W (n), (4)
length Lh . The OFDM receiver removes the LCP sam-
ples from the CP, and performs a NFFT -point FFT on the for n = 0, 1, ..., N − 1, with X(n) the nth data symbol, and
next NFFT received samples. The UFMC receiver applies a W (n) and H(n) the NFFT -point FFT of w and h respectively.
2NFFT -point FFT to the NFFT + L − 1 received samples We note that there is no interference term present in (4),
(padded with zeroes at the end) which correspond to the because the subcarrier signals are perfectly orthogonal and
UFMC symbol. At the receiver side, only the even-numbered ISI is not present, when LCP ≥ Lh − 1. At the receiver
−1
outputs of the 2NFFT -point FFT are kept, which are given by side, Y (n) is scaled
−1 with H(n) in case of OFDM and with
NFFT +L−2
H(n)Qi(n) (n) in case of UFMC.
1  −j2π Nmn
Y (n) = √ y(m)e FFT (2) The block diagrams of the systems for UFMC and OFDM
NFFT m=0 are depicted in figures 1 and 2, respectively.
for n ∈ {0, 1, ..., N − 1}, with y(m) denoting the received The adaptive coding and modulation (AMC) block at the
UFMC signal. transmitter looks for the optimal transmission mode (TM)
In the absence of CFO, Y (n) in case of UFMC can be vector Φ∗ = (r∗ , m∗ ), with m the bitloading vector and r
written as the code rate, which maximizes the GP. The GP, expressing
the number of useful information bits correctly decoded by
˜ + W̃ (n),
Y (n) = H(n)Qi(n) (n)X(n) + I(n) (3) the receiver per unit of time and frequency, is a representative
link performance metric. As the exact GP is a complicated
for n ∈ {0, 1, ..., 2N − 1}, where Qi (n) and H(n) are the
function of the SNRs on the individual subcarriers, we will
nth sample of the N -point FFT of qi and h, i(n) = n/D
instead perform the AMC by maximizing the estimated GP
is the subband index containing the subcarrier with index
(EGP), which can be expressed in terms of a scalar effective
n, and W̃ (n) is the 2nth sample of the 2NFFT -point FFT
SNR.
of the NFFT + L − 1 time-domain noise samples w with
w ∼ Nc (0, N0 IN ). The term I(n)˜ represents the interference For UFMC, the SNR associated with the nth subcarrier is
caused by the dispersive channel. The interference consists of given by
2
inter symbol interference (ISI), due to the dispersive channel, |H(n)|2 |Qi(n) (n)|
and inter carrier interference (ICI) due to the non-orthogonality γ(n) = (5)
N0
with i(n) denoting the index of the subband which contains The resulting 2nth FFT output Y (n) can be represented as
the nth subcarrier. Neglecting the interference in (3), the EGP the sum of a linear combination of the symbols X(n ) and a
is expressed as noise contribution W̃ (n). Substituting y(m) from (8) in (2),
it can be shown after some computations that
NFFT Np
ζ(Φ, U ) = · [1 − θr (γeff )] (6) NF
F T −1 NFFT
−1
NFFT + L − 1 U · N
Y (n) = X(n ) Qi(n ) (l)H(l)·
with U = r NN−1 u
the number or multicarrier symbols re- n =0 l=0
n=0 mn
quired to send one packet and Φ the TM. The quantity θr (γeff ) F (n − l − )F (n − n − ) (10)
is the packet error rate (PER) on an AWGN channel for BPSK
mapping and code rate r, when operating at an SNR equal to for n = 0, 1, ..., N − 1, and
γeff , which is the effective SNR of the multicarrier system.
We calculate γeff according to the κESM link performance 1 − exp(−j2πu)
F (u) =  
prediction method from [11], which gives γeff as a function u
NFFT · 1 − exp −j2π NFFT
of the SNRs on the individual subcarriers:
 N −1
 Denoting by A(n, n , ) the coefficient of X(n ) in Y (n), we
1  define
γeff = −log N −1 α(mn )e−γ(n)β(mn ) (7)
n=0 mn n=0

where α(mn ) and β(mn ) are known functions of mn . 2N


FFT −1

In the case of OFDM, the EGP is obtained by replacing A(n, n , ) = Qi(n ) (l)H(l)F (n − l − )F (n − n − )
in (6) L − 1 by LCP ; γeff is still given by (7), but γ(n) is l=0
computed from (5) with Qi(n) (n) = 1.
Note that A(n, n , 0) = Qi(n) (n)H(n)δn,n , with δn,n denot-
III. E FFECT OF CFO ing the Kronecker delta function, which is consistent with (3).
In the previous section, we considered zero CFO, in which The SINR associated with the nth subcarrier is given by
case the interference is zero (OFDM) or negligible (UFMC).
However, accurate synchronization involves a high complex- |A(n, n, )|2
SINRUFMC (n; ) =  (11)
ity; often a nonzero residual CFO is still present in the received N0 + n =n |A(n, n , )|2
signal [14]. In the case of OFDM, the performance degrades
very rapidly with increasing CFO, because of the increasing Applying a similar reasoning to OFDM, we obtain that [3]
interference power due to loss of subcarrier orthogonality.
UFMC is more robust against CFO because its subband |B(n, n, )|2
SINROFDM (n; ) =  (12)
filtering reduces the interference from out of subband carriers N0 + n =n |B(n, n , )|2
[15]. However, even with UFMC, the performance degradation
is not negligible for increasing values of the CFO. To improve with B(n, n , ) = H(n ) · F (n − n − ), the coefficient of
the GP performance in the presence of CFO, we propose X(n ) in Y (n); note that B(n, n , 0) = H(n)δn,n , which
to take the interference due to the CFO into account when agrees with (4). For OFDM, the SINR decreases faster for
performing the AMC; more specifically, we will use the SINR increasing , compared to UFMC.
(rather than the SNR) on the subcarriers when computing the In order to use SINRUFMC (k; ) or SINROFDM (k; ) in-
EGP, on which the AMC is based. stead of the SNR γ(k) for performing AMC, we need perfect
Therefore, we first calculate the interference power in a channel state information at the transmitter (CSIT), i.e., Qi (n),
multicarrier symbol. The ISI due to the dispersive channel H(n) and must be known. However, in practice typically
is neglected further on, since the subband filtering serves as only Qi (n) and H(n) are known, but not . Assuming that
a protection against the multipath channel [14]. For UFMC, the residual CFO after synchronization has a zero mean
we can write the received time-domain signal after imperfect Gaussian distribution with variance σ2 which is known to
CFO correction as the transmitter, we compute the SINR for the nth subcarrier
 NFFT
B−1 −1 by replacing in (11) |A(n, n , )|2 by E [|A(n, n , )|2 ] for
y(m) = c(m; ) xi (k)htot,i (m − k) + w(m) (8) all (n, n ), where E [.] denotes expecation with respect to
i=0 k=0 . We use the SINR resulting from (7) for calculating the
for m ∈ {0, 1, ..., NF F T + L − 2}, with htot,i the convolution EGP, which is the objective function to be maximized by the
of qi and h, and AMC algorithm. Similarly, in the case of OFDM we replace
in (12) |B(n, n , )|2 by E [|B(n, n , )|2 ] for all (n, n ). For
c(m; ) = ej2πm/NFFT . (9)
both multicarrier modulations, an increasing σ2 not only yields
with denoting the CFO normalized to the subcarrier spacing. more interference power, but also reduces the useful signal
At the receiver side a 2NFFT - point FFT is applied to y(m). power.
System parameter Value
# information bits (Np ) 1024 2.5
# CRC bits (NCRC ) 32
# subcarriers (N ) 120 OFDM, traditional AMC
# subbands (B) 10 OFDM, new AMC
FFT size (NFFT ) 1024 UFMC, traditional AMC
2
UFMC, new AMC
Subcarrier spacing (Δ) 10 MHz/1024
# bits per subcarrier (Dm ) {2,4,6}
code rates (Dr ) {1/2,2/3,3/4,5/6}
Distance transmitter-receiver 141 m 1.5
Cyclic prefic length (LCP ) 72

GP
Subband filter length (L) 73
1
TABLE I
S YSTEM PARAMETERS

0.5

IV. N UMERICAL R ESULTS 0


0 0.02 0.04 0.06 0.08 0.1
In this section, a comparison is made of the AMC al- σ2
gorithms, using either the SNR (ignoring the CFO) or the
Fig. 3. GP for both OFDM and UFMC for both AMC strategies when SNR0
SINR (using the distribution of the CFO), for both OFDM = 15 dB.
and UFMC. The comparion is made in terms of the actual
GP, obtained from a simulation consisting of transmitting
and decoding a large number of independent packets and 2.5
adding the number of information bits in the correctly decoded OFDM, traditional AMC
OFDM, new AMC
packets. In the simulations, the extended vehicular A (EVA) UFMC, traditional AMC
channel model [12] and the modified COST231 Hata path loss 2
UFMC, new AMC
model from [13] are implemented considering independent
channel realizations for each transmitted packet. The system 1.5
parameters are summarized in Table I. Note that we have taken
GP

L = LCP + 1, since it has been claimed that the optimal


subband filter length for UFMC is approximately the same as 1
the CP length in OFDM system, and this allows a coexistance
of the two multicarrier waveforms [14]. For further use, we
define SNR0 = E[|X|2 ]/N0 . 0.5

Figure 3 shows the actual GP as a function of σ2 for OFDM


and UFMC, with AMC based on either the SNR (referred to 0
as traditional AMC) or the average SINR (referred to as new 0 0.02 0.04 0.06 0.08 0.1

AMC), at SNR0 = 15 dB. For both AMC strategies and both
multicarrier modulations, the GP decreased with increasing
Fig. 4. GP for varying  for both AMC strategies when new AMC method
σ2 . For a given AMC strategy, UFMC outperforms OFDM; uses σ2 = 0.025, and SNR0 = 15 dB.
for a given multicarrier format, the new AMC performs better
than the traditional AMC. For large σ2 , the GP for OFDM
with traditional AMC gets close to zero; OFDM with the new the traditional AMC for the given σ2 is valid for the simulated
AMC and UFMC with traditional AMC yield about the same SNR0 range from 5 to 25 dB.
GP, but a substantial gain is achieved for UFMC with the new
AMC. V. C ONCLUSION
In figure 4 the actual GP is displayed as a function of the The traditional AMC, based on the SNRs of the subcarriers
actual value of , with the new AMC method using σ2 = of OFDM and UFMC systems, yields a poor GP performance
0.025. For small , the GP performance for the new AMC in the presence of a CFO, due to the interference caused by
is worse compared to the traditional AMC, because the latter the latter. We have achieved a substantial increase of the GP
assumes = 0. However, already for > 0.001 the new AMC by applying a new AMC method, based on the average SINR
method outperforms the traditional AMC method. (rather than the SNR) of each subcarrier. Compared to the
Figure 5 shows the actual GP for a zero-mean Gaussian traditional AMC method, the new AMC reduces the code rate
with variance σ2 = 0.025 as a function of SNR0 . For OFDM and constellation sizes according to the average interference
with the traditional AMC, the GP is almost independent of power obtained from the statistical distribution of the CFO,
SNR0 , due to the dominant effect of the interference compared thereby achieving a lower packet error rate and substantially
to the noise power N0 . The gain of the new AMC method over increasing the GP.
[14] L. Zhang, A. Ijaz, P. Xiao, M. Imran, R. Tafazolli, “MU-UFMC
1.8 System Performance Analysis and Optimal Filter Length and Zero
OFDM, traditional AMC Padding Length Design”arXiv:1603.09169 [cs.IT], Submitted to IEEE
1.6 OFDM, new AMC Transactions on Signal Processing
UFMC, traditional AMC [15] A. Aminjavaheri, A. Farhang, A. RezazadehReyhani, B. Farhang-
1.4 UFMC, new AMC Boroujeny, “Impact of timing and frequency offsets on multicarrier
waveform candidates for 5G”, IEEE Signal Processing and Signal
1.2 Processing Education Workshop (SP/SPE), Aug. 2015

1
GP

0.8

0.6

0.4

0.2

0
5 10 15 20 25
SNR0 [dB]

Fig. 5. GP for σ2 = 0.025 and varying SNR.

ACKNOWLEDGMENT
This research has been funded by the Interuniversity Attrac-
tion Poles Programme initiated by the Belgian Science Policy
Office.

R EFERENCES
[1] X. Wang, T. Wild, F. Schaich, S. ten Brink, “Pilot-aided Channel
Estimation for Universal Filtered Multi-Carrier”, Vehicular Technology
Conference (VTC Fall), Sept. 2015.
[2] X. Wang, T. Wild, F. Schaich, A. Fonseca Dos Santos, “Universal Fil-
tered Multi-Carrier with Leakage-Bssed Filter Optimization”, European
Wireless Conf., 2014.
[3] B. Xie, W. Qiu, H. Minn “Exact Signal Model and New Carrier
Frequency Offset Compensation Scheme for OFDM” IEEE Trans. on
Wireless Comm., Vol. 11, No.2, pp. 550-555, Febr. 2012.
[4] G. Wunder et al. “5G NOW: Non-Orthogonal, Asynchronous Waveforms
for Future Mobile applications”, IEEE Comm. Mag., vol. 52, no. 2, pp.
97-105, Febr. 2014.
[5] X. Wang, T. Wild, F. Shaich “Filter Optimization for Carrier-Frequency-
and Timing-Offset in Universal Filtered Multi-Carrier Systems” Vehic-
ular Tech. Conference (VTC Spring), May 2015.
[6] C. Vitiello, P. Del Fiorentino, E. Debels, V. Lottici, F. Gianetti, M.
Luise, M. Moeneclaey, “ Two-Step Resource Allocation for BIC-UFMC
Wireless Communications”, presented to ISWCS 2016
[7] V. Vakilian, T. Wild, F. Schaich, S. ten Brink, J. Frigon, “Universal
Filtered Multi-Carrier Technique for Wireless Systems Beyond LTE”,
9th Int. Workshop on Broadband Wireless Acces, Globecom, Dec. 2013
[8] M. Mukherjee, L. Shu, P. Kumar, R. Matam, “Reduced out-of-band
radiation-based filter optimization for UFMC systems in 5G”, Wireless
Comm. and Mobile Computing Conference (IWCMC), 2015..
[9] P. Del Fiorentino, C. Vitiello, V. Lottici, F. Gianetti, M. Luise, E.
Debels, J. Van Hecke, M. Moeneclaey, “Resource Allocation in Short
Packets BIC-UFMC Transmission for Internet of Things”, Accepted to
GLOBECOM 2016
[10] I. Stupia, F. Giannetti, V. Lottici, L. Vandendorpe, “A Greedy Algoritm
for Goodput-Based Adaptive Modulation and Coding in BIC-OFDM
Systems”, European Wireless Conf. , pp. 608-615, April 2010.
[11] I. Stupia, V. Lottici, F. Giannetti, L. Vandendorpe, “Link Resource Adap-
tation for Multiantenna Bit-Interleaved Coded Multicarrier Systems”,
IEEE Trans. on Signal Proc., vol. 60, no., pp 3644-2656 July 2012
[12] Extended vehicular A (EVA) channel. Available:
http://www.raymaps.com/index.php/lte-multipath-channel-models.
[13] IEEE 802.16m Evaluation Methodology Document (EMD), “IEEE
802.16m-08/004r2”, pp 143, July 2008

You might also like