Sei sulla pagina 1di 5

European Wireless 2014

Universal Filtered Multi-Carrier with


Leakage-Based Filter Optimization
Xiaojie Wang * , Thorsten Wild , Frank Schaich , and Andre Fonseca Dos Santos

*
University of Stuttgart, Email: gdi73302@gmail.com
Alcatel-Lucent
Bell Labs, Lorenzstr. 10, 70435, Stuttgart, Germany
Email: {thorsten.wild, frank.schaich, a.santos}@alcatel-lucent.com

Abstract Universal Filtered Multi-Carrier (UFMC) is a novel kinds of MIMO. UFMC can also rely on FFT-based receive
modulation technique which can be seen as a generalization processing with per-subcarrier equalization. This paper focuses
of ltered OFDM and lter bank based multi-carrier (FBMC- on the lter optimization for UFMC. While previous work
FMT). Being a candidate waveform technology for 5G wireless
systems it combines the simplicity of OFDM with the advantages [3], targeting the coordinated multi-point (CoMP) application
of FBMC. While previous work on UFMC was based on Dolph- scenario, uses Dolph-Chebyshev lters as a reasonable ad-hoc
Chebyshev-lters, in this paper we now optimize the FIR lter choice, we now want to optimize the lter coefcients. Two
design, using two different optimization criteria. The rst crite- different optimization criteria are introduced and discussed.
rion is the maximization of the signal over out-of-band leakage UFMC, based on those criteria, is then compared in simula-
ratio. The second criterion uses signal over in-band distortion
plus out-of-band leakage ratio. We compare UFMC based on tions against UFMC with Dolph-Chebyshev lters and against
lters with those two optimization approaches against UFMC classical CP-OFDM, showing signicant gains in terms of SIR,
with Dolph-Chebyshev lters and against OFDM in a setting when carrier frequency offset (CFO) is present.
with non-perfect frequency alignment. The performance gains of The remaining part of this paper are structured as follows:
the optimized lters are demonstrated in link level simulations. Section II describes the system model and analyzes the impact
Both of the proposed approaches show signicant performance
gains over OFDM in terms of average SIR: Almost 10 dB and of CFO to UFMC. Section III introduces the optimization
7 dB SIR improvement can be achieved by using the rst and criteria to design FIR-lter. Section IV shows performance re-
second criterion, respectively. sults, based on simulations. Section V concludes and provides
the outlook.

I. I NTRODUCTION II. S YSTEM M ODEL


Todays most prominent multi-carrier modulation technique
is OFDM, being used in standards like LTE and WiFi. OFDM
is very simple and efcient with FFT-based modulation and de-
modulation and can rely on per-subcarrier scalar equalization.
Its main drawbacks are the high spectral side-lobe levels, due
to the rectangular symbol shape in time. This makes OFDM
vulnerable against time-frequency misalignments. In practical
usage, OFDM is usually ltered in order to ensure proper
out-of-band radiation. A future 5G candidate technology for
replacing OFDM is lter-bank based multi-carrier (FBMC)
[1]. Each subcarrier is individually ltered, strongly enhancing
robustness against inter-carrier interference (ICI) effects. A
typical FBMC drawback is its long lter length, making it
disadvantageous for communication in short uplink bursts, as
required in potential application scenarios of 5G systems [2],
like low latency communication or energy-efcient machine-
type communication (MTC). Universal Filtered Multi-Carrier
(UFMC) is a novel modulation technique [3], which is a
generalization of ltered OFDM and FBMC (in its ltered Fig. 1. Simplied system model of UFMC under consideration of CFO
without channel
multi-tone (FMT) variant). While the former lters the entire
band and the latter lters each subcarrier individually, UFMC
lters subband-blocks, thus groups of subcarriers. This allows Consider a simplied UFMC system, shown in Fig. 1, in
reducing the lter length considerably, compared to FBMC. which the overall subcarriers N are divided into B subbands.
Furthermore, QAM modulation is still efcient (in contrast A subband in UFMC may also correspond to a physical re-
to the FBMC case [1]), making UFMC compatible to all source block (PRB) in LTE. Each subband has NB consecutive

ISBN 978-3-8007-3621-8 VDE VERLAG GMBH, Berlin, Offenbach, Germany


963
European Wireless 2014

subcarriers. The frequency domain subband signal Xi with an In order to separate the signal part from the interference part
input length of NB is converted into the time domain signal in (5) we rstly derive Xi from (1) as follows
xi with an output length of N by N -point IDFT spreader. k

Xi 2 if k is even
 
After IDFT operation in each subband, the output signal is

ltered by a FIR-lter with the length of L. The transmitted
sin 2 (2m k)
Xi (m)  

signal of UFMC is then the sum of all ltered subband signals. Xi (k) = mSi N sin 2N (2m k) (7)


Because of convolution with the FIR-lter, a UFMC symbol
j
e 2

(2mk)(1 1
)
has the length of N + L 1. Thus after appending zeros a

N

2N point FFT is to be performed at the receiver, in order to if k is odd
estimate the symbols in individual subcarriers. The transmitter where Xi is the frequency domain signal of the i-th subband
and receiver local oscillator misalignment and Doppler effect and can be written as
due to moving users are considered and modeled as CFO.
0 if k
/ Si
For an arbitrary subband with the index i, the signal in Xi (k) = (8)
Xi (k) if k Si
frequency domain Xi is converted into time domain xi by
IDFT and can be written as follows: As we can observe from (7), all the odd subcarriers contain
1  part of signal energy and ICI from other subcarriers in the
xi (l) = Xi (k)ej2kl/N , l = 0, ..., N 1; (1) subband because of the 2N -point FFT based detection. For
N
kSi
simplicity reason, we rewrite (7) for a considered subcarrier
where Si is a nonempty set, which contains the subcarrier k in subband i as
indexes in the subband i. Due to ltering, the output signal yi
can be expressed as a discrete linear convolution between the Xi,k (m) = Xi,Sk (m) + Xi,ICIk (m) (9)
FIR-lter fi and the time domain signal xi . Xi,Sk (m) denote the spread signal in subcarrier m from the
yi = x i f i (2) subcarrier k and Xi,ICIk (m) is the corresponding ICI. Only
even subcarriers are of interest, thus k = 0, 2, , 2N 2.
where denote the linear convolution operator. And as pre- But due to convolution, odd subcarriers must be considered
viously already mentioned yi has the length of N + L 1. to obtain the symbol estimates in subcarriers k. Thus, m =
Therefore, the UFMC symbol y under consideration of CFO 0, 1, , 2N 1. They are straightforward with (7)
can be expressed as m
Xi 2


if m = k
X  k  sin( 2 (km)) ej 2 (km)(1 N1 )
B

y = 2N ci (xi fi ) (3) i 2 N sin (km)
( 2N )
Xi,Sk (m) =
i=1
if m odd


ci is the time domain expression of frequency offset for the
0 otherwise
i-th subband and can be written as (10)
1 j2i k/N and similarly
ci (k) = e , k = 0, ..., N + L 2; (4)
2N

0, if m even
where i denote relative CFO (rCFO) for subband i. rCFO
  

sin 2 (2l m)
means that CFO is normalized to the subcarrier spacing. Since
Xi (l)  

we focus on the the effect of rCFO, the channel is assumed to Xi,ICIk (m) = lSi ,l= 2 k
N sin 2N (2l m)


be perfect. At the receiver, the 2N-point FFT based detection is
ej 2 (2lm)(1 N )
1


used to convert the received time domain signal into frequency

domain. Without loss of generality, we can assume that the if m is odd
estimated symbol in the subcarrier k belongs to the subband (11)
i. Then we can derive the estimated symbols as follows With (9) to (11), we are able to separate the signal part from
the interference part in the estimated symbols in (5).
2N
 1
Y (k) = Ci (k j)Xi (j)Fi (j) 2N
 1

j=0 Y (k) = Ci (k j)Xi,Sk (j)Fi (j) . . . Signal


B 2N 1
(5) j=0
 
+ Cl (k j)Xl (j)Fl (j) 2N
 1

l=1,l=i j=0 + Ci (k j)Xi,ICIk (j)Fi (j) . . . ICI


j=0
where Ci , Xi and Fi are 2N-point FFT of ci , xi and fi B 2N 1
 
respectively. And they are assumed to be periodic with period + Cl (k j)Xl (j)Fl (j) . . . IBI
of 2N . Furthermore, from (4), Ci can be expressed as l=1,l=i j=0
  (12)
sin 2N (2i k) (N + L 1)
Ci (k) =  
2N sin 2N (2i k) (6) As the above equation (12) shows, the estimated symbol gets

j 2N (2i k)(N +L2) distortion in case of CFO both from subcarriers in subband,
e

ISBN 978-3-8007-3621-8 VDE VERLAG GMBH, Berlin, Offenbach, Germany


964
European Wireless 2014

 
which we call ICI here, and also from all other subbands, where S = Sk and ZOBL = ZOBL,k . Since only
which we call Inter Band Interference (IBI) here. Due to k/2Si k/2S
/ i
2N -point FFT additionally, the spread signal part in all odd even subcarriers are of interest, k = 0, 2, , 2N 2. Hence
subcarriers are weighted and added together to form the nal the optimization task can be formulated as:
signal part. fo = arg max SLR
f (18)
III. C ONSIDERED O PTIMIZATION C RITERIA subject to f 2 = 1
In comparison to OFDM, UFMC introduces L lter coef- It was shown in [4] that the solution for (18) above is given
cients, which can be designed to gain more robustness against by  
CFO. Direct maximization of Signal to Interference plus Noise fo max. eigenvector Z1 OBL S (19)
Ratio (SINR) generally results in a challenging optimization
with coupled variables [4], especially if different subbands use the eigenvector corresponding to the largest eigenvalue of the
different lters. Now we consider only one subband so that IBI matrix Z1OBL S, if the matrix ZOBL is invertible. In general
disappears and (12) can be simplied as the solution is given by [5]
2N
 1 fo max. generalized eigenvector (ZOBL , S) (20)
Y (k) = Ci (k j)Xi,Sk (j)Fi (j)
j=0
The numerator and denominator of (17) represent the sum of
2N 1
(13) signal energy in subband and the sum of the energy of OBL

+ Ci (k j)Xi,ICIk (j)Fi (j) into other subbands respectively.
j=0 In the following we carry out the optimization and show
the optimized lter in time and frequency domain compared
Observing the equation above, we can discover that the
to Dolph-Chebyshev lter with 40 dB side lobe attenuation,
CFO causes in-band distortion (IBD) and out-of-band leakage
which was used in our previous work. The optimization is
(OBL), which correspond to ICI and IBI in (12). For sub-
carried out for lter length L = 16, FFT-Length N = 128 and
carriers in the subband of interest the rst term is the signal
subband size NB = 12.
YS (k) and the second term is IBD YIBD (k). For subcarriers
outside of subband of interest the rst term disappears and In Fig. 2 the lters are optimized for

deterministic rCFOs
the second term is OBL YOBL (k), which causes distortion to and the lter energy is constrained to fi2 = 1. As we can
another subband for subcarriers, which are allocated there. see from the gure, the results remain the same for positive
Now we derive the formula of expected signal energy, IBD and negative rCFO. Furthermore, this criterion leads to equal
and OBL energy. For simplicity reason, we drop the subband
index i. Firstly we assume that the symbols in different sub-
band subcarriers are uncorrelated E{Xi (l)Xi (k)} = 0, if l =
k and the symbol energy is normalized to E{Xi (k)Xi (k)} =
1. Then we are also able to formulate the expected signal
energy Sk , IBD energy ZIBD,k and OBL energy ZOBL,k for
subcarrier k in matrix form as follows
Sk = E{YS (k)YS (k)} = f H Sk f . (14)
and similarly

ZIBD,k = E{YIBD (k)YIBD (k)} = f H ZIBD,k f . (15)

ZOBL,k = E{YOBL (k)YOBL (k)} = f H ZOBL,k f . (16)
Fig. 2. SLR optimized lter for deterministic rCFO normalized to subcarrier
where ()H means Hermitian transpose, () is the conjugate spacing

complex operator and f is a vector, containing all the lter


coefcients in time domain. As a result, f has the dimension
ripple lter design approach and the optimized lters closely
of L 1 and Sk , ZIBD,k and ZOBL,k are of the dimension
resemble Dolph-Chebyshev lters with different side lobe
L L.
attenuations. For example, the lter optimized for rCFO= 0.1
In the following we propose two methods to design the
has almost the same characteristic in frequency domain as
FIR-lter for UFMC which results in a generalized eigenvalue
Dolph-Chebyshev lter with side lobe attenuation of 13.4 dB.
problem similar as in [4].
While the rCFO increases, the optimized lter has a smaller
main lobe width and side lobe attenuation.
A. maximizing SLR
Instead of deterministic rCFO, Fig. 3 shows the result
In this approach we dene the so-called Signal to out-of- for a uniformly distributed rCFO in the range of 0.5 to
band Leakage Ratio (SLR) as 0.5 subcarrier spacing. In this case the optimized lter can
fHS f be approximated by Dolph-Chebyshev lter with side lobe
SLR = (17)
f H ZOBL f attenuation of 12.1 dB.

ISBN 978-3-8007-3621-8 VDE VERLAG GMBH, Berlin, Offenbach, Germany


965
European Wireless 2014

Fig. 3. SLR optimized lter for uniformly distributed rCFO between 0.5 Fig. 5. SDLR optimized lter for uniformly distributed rCFO between 0.5
subcarrier spacing subcarrier spacing

approach maximizing SLR, the optimized lters are not equal


B. maximizing SDLR ripple anymore and show small variation for different rCFOs.
Both of the two proposed lter design approaches does not
Similar as the rst method SLR, we also take IBD into
require any other knowledge of other subbands and have a
account in the second proposal. And the Signal to in-band
closed-form solution. Although we neglect IBI from other
Distortion plus out-of-band Leakage Ratio (SDLR) is dened
subbands in the optimization, if all the subbands use these
as
fHS f criteria to minimize OBL, the IBI is also reduced.
SDLR = H (21)
f (ZIBD + ZOBL ) f
 IV. R ESULTS
where ZIBD = ZIBD,k . Similarly the optimization task
Now the optimized lter in previous section is to be used
k/2Si
is then: in an UFMC system with FFT size N = 128, lter length
fo = arg max SDLR L = 16, subband size LB = 12 and total number of subbands
f (22) B = 10. QPSK is used for data symbol modulation. No guard
subject to f 2 = 1
subcarriers are used between subbands and all subbands use
The solution is similar as (19). In this approach, the denomina- the same type of lter, except that the lters are shifted to the
tor represents the sum of interference energy both in subband center frequency of each subband. Simulations are carried out
of interest and out of subband of interest. Additionally we nd for two different scenarios. In scenario 1 the UFMC signal
that the matrix ZIBD,k is a scalar multiple of Sk . is distorted by CFO for all subbands. So every subcarrier and
subband suffer from ICI and IBI. This corresponds to the case,
Sk = k ZIBD,k (23)
that the CFO remains untreated by the receiver. In scenario 2
which means the Signal over in-band Distortion Ratio the UFMC signal is distorted by CFO for all subbands, except
(SIBDR) for subcarrier k is k , independent of lter coef- that one subband (User of Interest,UoI) is perfect synchronized
cients. Therefore we can conclude that the IBD cannot be in receiver. This corresponds to a scenario, where users are
minimized by ltering, because the SIBDR is independent of not perfectly aligned in terms of CFO, but the receiver is able
lter coefcients. to estimate and compensate the CFO. In this case there is
In the Fig. 4 and 5 the optimized lters are shown for no ICI coming from within the subband anymore. Distortions
deterministic rCFOs and uniformly distributed rCFO between are OBL from other subbands. The corresponding SIR are
0.5 subcarrier spacing respectively. In contrast to the rst quantied for different rCFO and system settings.
In Fig. 6 the SIR are shown for all the used subcarriers in
the presence of rCFO= 0.1. The used lters are optimized
for this deterministic rCFO. According to the simulation
results, UFMC is generally more robust against CFO, since
by designing the lter, UFMC has less leakage energy in
neighbor subcarriers than OFDM in case of CFO. The SIR
is almost identical for all the used subcarriers for OFDM,
except for subcarriers at the edge. The reason is that not all
the subcarriers are used for data transmission, so that the edge
subcarriers get less interference. Furthermore, the simulated
average SIR for OFDM is 14.74 dB, which is almost the same
as theoretical SIR [6]. Moreover, the observed SIR for UFMC
characteristic varies within the subcarriers of the respective
Fig. 4. SDLR optimized lter for deterministic rCFO normalized to subcarrier
spacing
subbands. This variation repeats for each subband. In addition,
the subcarriers in the middle of the subband have a lower

ISBN 978-3-8007-3621-8 VDE VERLAG GMBH, Berlin, Offenbach, Germany


966
European Wireless 2014

Fig. 6. Scenario 1: Performance comparison for rCFO=0.1


Fig. 8. Performance comparison for different rCFO

SIR value than other subcarriers in the edge. In this scenario V. S UMMARY
UFMC with a Dolph-Chebyshev lter with 40 dB side lobe
attenuation gains 0.6 dB in terms of average SIR over OFDM. In this paper we have analyzed a UFMC system in the pres-
And by using max. SDLR approach UFMC gets 0.72 dB gain. ence of CFO and proposed a leakage based lter optimization,
Greatest gain is 0.83 dB by using max. SLR. Interesting is that maximizing SLR and SDLR, to improve robustness against
SLR outperforms SDLR and the gain is not very signicant. CFO for UFMC. These lter design approaches have closed-
form solution and lead to a generalized eigenvalue problem.
For the second scenario in the presence of deterministic With these criteria we are able to design lters independently
rCFO= 0.1, only distortion from other subbands to UoI is without any knowledge of other subbands. From the simulation
considered. The SIR for subcarriers in the UoI allocation are results, max. SLR approach outperforms max. SDLR, since
shown in Fig. 7. In contrast to scenario 1, the edge subcarriers the SDLR maximization is limited by the fact that the in-band
distortion is a scalar multiple of useful signal, independent
of the lter coefcients. Signicant gains appear for the case
that the user of interest is perfect synchronized, still resulting
into inter-carrier interference to frequency-misaligned adjacent
channel users. This means, through the proposed lter design
approach, the distortion to other subbands is signicantly
reduced if the considered subband is distorted by CFO.

ACKNOWLEDGMENT
A part of the research leading to these results has received
funding from the European Communitys Seventh Frame-
work Program [FP7/2007-2013] under the grant agreement
n 318555 referred as 5GNOW.
Fig. 7. Scenario 2: Performance comparison for rCFO=0.1
R EFERENCES
[1] B. Farhang-Boroujeny; OFDM versus lter bank multicarrier, IEEE
have now lowest SIR value. The reason is that IBD is not Signal Process. Mag., vol. 28, pp. 92 112, May 2011.
present anymore and the edge subcarriers suffer more OBL [2] Wunder, G.; Kasparick, M.; ten Brink, S.; Schaich, F.; Wild, T.; Gaspar, I.;
Ohlmer, E.; Krone, S.; Michailow, N.; Navarro, A.; Fettweis, G.; Ktenas,
from other subbands than those in the middle. In this case, D.; Berg, V.; Dryjanski, M.; Peitrzyk, S.; Eged, B.; 5GNOW: Chal-
UFMC with Dolph-Chebyshev lter with side lobe attenuation lenging the LTE Design Paradigms of Orthogonality and Synchronicity,
of 40 dB has a SIR gain of 4.62 dB over OFDM averaged over Mobile and Wireless Commun. Syst. for 2020 and Beyond, Workshop @
77th IEEE Veh. Technol. Conf. Spring (VTC13 Spring), Jun. 2013.
the subband. And by using max. SDLR and max. SLR, SIR [3] Vakilian,V.; Wild, T.; Schaich, F.; ten Brink, S.; Frigon, J.-F.; Universal
gain of 6.81 dB and 9.57 dB can be achieved respectively. Filtered Multi-Carrier Technique for Wireless Systems Beyond LTE,
9th Int. Workshop on Broadband Wireless Access at IEEE Globecom13,
Similarly, the lter optimized for uniformly distributed Atlanta, Dec. 2013.
rCFO between 0.5 and 0.5 subcarrier spacing is used for [4] Mirette Sadek; Student Member; IEEE; Alireza T.; Ali H. Sayed; A
scenario 1 and 2, and the corresponding resulting gains over Leakage-Based Precoding Scheme for Downlink Multi-User MIMO
Channels, IEEE Transactions on Wireless Communications, Vol. 6, No.
OFDM are shown in Fig. 8 for rCFOs in this range. The 5, May 2007.
performance is symmetric for positive and negative rCFOs, [5] M. Sadek; A. Tarighat; A. H. Sayed; Active antenna selection in multi-
so we only show the SIR gain for positive rCFOs. Obviously user MIMO communications,IEEE Trans. Signal Processing, Vol. 55,
No. 4, pp. 1498-1510, April 2007.
UFMC is more robust against CFO than OFDM and the SLR [6] Jungwon Lee; Huiling Lou; Dimitris Toumpakaris; John M. Ciof; Effect
approach outperforms SDLR in both scenarios. The SIR gain of Carrier Frequency Offset on OFDM Systems for Multipath Fading
over OFDM decreases, while the rCFO increases. Channels,Global Telecommunications Conference, 2004.

ISBN 978-3-8007-3621-8 VDE VERLAG GMBH, Berlin, Offenbach, Germany


967

Potrebbero piacerti anche