Sie sind auf Seite 1von 6

A Reduced Complexity Transmitter for UF-OFDM

Thorsten Wild and Frank Schaich

Bell Labs, Alcatel-Lucent, Germany


Email: {thorsten.wild, frank.schaich}@alcatel-lucent.com

AbstractUF-OFDM is a promising 5G waveform candidate, low-end devices) and allows for a kind of connection-less
close to CP-OFDM, but with better spectral properties. Efficient operation mode. Furthermore, the improved spectral separation
receiver implementations exist, which are close to CP-OFDM in among subbands allows for the usage of different waveform
terms of complexity. The UF-OFDM multi-carrier modulator, on
the other hand, still has no efficient solution in the literature parameters in parallel subbands, e.g. for supporting on a per-
up to now. This paper addresses this problem and presents user basis special modes for low latency or high velocity.
a novel frequency domain generation method for UF-OFDM, The multi-carrier receiver complexity of UF-OFDM is mod-
where overlapping subbands are superimposed. Our approach est: Slightly higher than factor 2 of CP-OFDM, as a 2-N
can bring down UF-OFDM multi-carrier modulator complexity
almost as low as a factor of 2 above CP-OFDM, with a negligible
point FFT is used for subcarrier demodulation. Regarding the
approximation error. Thus the required efficient transmitter UF-OFDM transmitter, so far only a matrix multiplication
implementation is delivered by this papers solution, ready to has been presented as the textbook-solution. This approach
be used in hardware implementations. scales
! badly
 in terms of real multiplications and additions with
O N 2 , compared to the nice O (N log N ) scaling of N -point
I. I NTRODUCTION
FFT implementations of CP-OFDM.
Todays dominating wireless physical layer waveform is This paper now addresses the multi-carrier modulator com-
cyclic-prefix (CP-)OFDM. It is used in fourth generation plexity issue of UF-OFDM. A novel solution is presented
cellular systems, such as LTE(-A), as well as in IEEE 802.11 which brings the complexity order down to O (N log N ), alike
standards. A clear strength of CP-OFDM is its simplicity: CP-OFDM. The exhibited approach generates the signal in
Transmitter and receiver can be realized by efficient fast frequency domain. Each subband is constructed with small
Fourier transforms (FFT), followed by simple scalar equal- IFFTs and FFTs (e.g. factor 16 smaller than the full FFT-size),
ization. A drawback of CP-OFDM is its spectral property of the filtering is executed as multiplication in the frequency
high side-lobe levels. As a consequence, CP-OFDM has to be domain and the overlapping subbands with their side-lobes
operated in strict time-frequency alignment in order to avoid are superimposed in a large 2-N point IFFT. Note that also
inter-carrier interference. in the FBMC community, methods for frequency domain
In research, fifth generation wireless systems (5G) are signal generation exist [9]. Here the main lobe of the filtered
already discussed, targeting the Horizon 2020 [1][2]. 5G subcarriers is synthesized. In UF-OFDM, as the filters are
will be designed for very heterogeneous service- and device per subband with some overlap, the signal generation is done
classes. In order to suit 5G needs better than CP-OFDM, new subband-wise in an overlap-add fashion.
waveform candidates are currently discussed [3]. All those As the proposed method is an approximative solution for
candidates have better spectral properties than CP-OFDM: UF-OFDM we carefully check the resulting distortions and
Filter-Bank Multi-Carrier (FBMC) [4], Generalized Frequency spectral properties. By proper parameter design, trading off
Division Multiplexing (GFDM) [5] and a new contender which complexity with approximation errors, we find the point where
is called Universal Filtered Multi-Carrier (UFMC) [6]. UFMC the error floor can be considered as negligible. The complexity
is in its nature very close to OFDM. Instead of a CP, it uses is computed and compared to CP-OFDM.
soft symbol transitions, caused by per-subband FIR filters
which filter groups of subcarriers. This approach thus has
much shorter filter lengths than FBMC which makes UFMC II. S YSTEM M ODEL
more suitable to short burst communication [6]. As UFMC is
very close in nature to OFDM it is also called UF-OFDM [7]. First the UF-OFDM basic multi-carrier modulator is de-
UF-OFDM, due to its better spectral properties has shown to picted. Then an approximation by the novel frequency domain
be much more robust against time-frequency misalignments generation method is described.
T
than CP-OFDM [8]. This is advantageous for integrated 5G Notation: We use () to denote transpose. ~0N M denotes
air interfaces for supporting synchronous broadband traffic the matrix of size N M with all zero entries, Expectation
in parallel to small packet services with relaxed synchronic- is denoted by E {}. All vectors are column vectors, denoted
ity, i.e. less strict time-frequency alignment. This relaxed in small bold x for time domain representation and capital
synchronicity helps to save signaling overhead and battery bold X for frequency domain representation. x y denotes
consumption, relaxes local oscillator requirements (helpful for the Hadamard product. All matrices are in capital bold F.

978-1-4799-8088-8/15/$31.00 2015 IEEE


frequency domain representation, upsampled by factor NOS
IDFT Filter F1k x1k
s1k spreader V1k with
which is typically chosen as NOS = 2, can be written as
+ P/S length L h iT 
T ~
Xi = FFT si , 0[1(NOS 1)N )] . (2)
IDFT spreader Filter F2k x2k
s2k V2k with xk Baseband i =
+ P/S length L + to RF upsampled frequency domain subband filter F
The 
h iT
T ~
FFT fi , 0[1NOS N L] can now carry out the filtering
by element-wise multiplication Fi X i . Summing up all the
IDFT spreader Filter FBk xBk
sBk VBk with channel different subband contributions yields
+ P/S length L
B
X
noise n other total =
X F i.
i X (3)
+ users
symbol i=1
estimates Time
domain
In order to obtain the time domain
samples we carry out an
Frequency pre- RF to
NOS N -point IFFT x total = N IFFT {Xtotal }. The UFMC
domain processing
symbol (e.g.
Baseband symbol consists of the first N + L 1 samples, the remaining
2N point-
processing FFT windowing) ones are zeros:
(e.g. per + S/P
subcarrier T
equalization) 0 xtotal (1), x
x = [ total (2), . . . , x
total (N + L 1)] (4)
0
zeropadding
The complexity of this approach will be discussed in section
0
III.B, where it becomes clear that complexity reductions are
required.
Fig. 1. Generic UF-OFDM baseline transceiver
C. Novel Frequency Domain Approximation
A. UF-OFDM Baseline Transmitter The idea for reducing the frequency domain implementation
complexity exploits the fact that subband signals have to be
The base-band time-domain transmit vector xk for a partic-
generated with a lower sample rate, compared to the full band.
ular multi-carrier symbol of user k is the superposition of the
Furthermore, the proposed technique cuts the signal in fre-
sub-band-wise filtered components, with filter length L and
quency and time to provide a low complex frequency domain
FFT length N , and can be represented by
signal approximation. The aim is to make the approximation
B
X error negligible, e.g. much smaller than any distortions caused
xk = Fi,k Vi,k si,k . (1) by the subsequent RF chain processing and digital RF pre-
|{z} |{z} |{z} |{z}
i=1
[(N +L1)1] [(N +L1)N ] [N ni ] [ni 1] processing.
For the i-th subband (1 i B), the data symbols In order to make the solution adjustable, with different
si,k Cni 1 are transformed to time-domain by the tall trade-offs for complexity and accuracy, we introduce tuning
IDFT-matrix Vi,k , which includes the relevant columns of the parameters. Small cut-off IFFTs are used to generate each
inverse Fourier matrix according to the respective sub-band subband in time domain. Besides the data-carrying pass-band
position within the overall available frequency range. Fi,k is part, they contain the relevant side-lobes to preserve the wave-
a Toeplitz matrix, composed of the FIR filter impulse response form characteristics. Those small IFFTs are thus overlapping.
fi , performing the linear convolution. Fi,k is a waveform de- We denote their size as NIF F T o . Each subband construction
sign parameter. It can be adjusted by the system to propagation is thus an upsampling of frequency domain QAM symbols
conditions and time-frequency offset requirements [10]. For si,k , followed by frequency domain filtering for generating
simplicity, in this paper we constrain the UF-OFDM waveform desired spectral properties. All the subbands are superimposed
to identical subband sizes ni = Q and the subband FIR filters in one large FFT with a size of NOS N , where NOS denotes
fi to frequency shifted versions of the same prototype filter f . the oversampling factor. The overall implementation tuning
parameters are thus NOS and NIF F T o . In our investigations
B. Baseline Frequency Domain Implementation NOS = 2 was large enough to sufficiently approximate the
A straightforward frequency domain implementation of signal, so section IV will focus on the variation of NIF F T o and
the transmit signal model (1) would consist of per-subband check which NIF F T o is required to make the approximation
operations of filtering and IFFT, summing up all the sub- solution accurate enough.
band outputs. This brute-force implementation choice is The approximated frequency domain overlap-add gen-
not recommended and just listed for comparison. The eration of UF-OFDM, illustrated by Fig. 2, is done in the
time domainsignal at the imput of subband filter i is subsequently described steps. For ease of notation, the user
h iT  index k and the subband index i is dropped. The data-
si = IFFT ~0[1(i1)Q] , sTi , ~0[1(N (i1)Q)] and its carrying frequency-domain input signal vector s is expanded
ers, f is shifted by half a subcarrier in frequency. The k-
xshort xext Fcut(1) th element of shift vector ~ is k = exp(j(k 1)N ).
0
0 This shifted filter is padded with zeros for a NOS N -FFT,
h iT
by FOS = FFT (~ f )T , ~0[1NOS N L] . Fcut is the
NIFFTopt. IFFT

0 Fcut(2)
appropriate cut-out for the small IFFT size NIF F T o . The
sik cut-off of frequency domain filter parts is one source of the
approximation error.
0
The filtered subband-component Xf ilt is now placed at

2Npt. IFFT
NFFTupt. FFT

Xfull(i) its respective frequency position in the large NOS N -point


0
Xfilt FFT. Its index offset is Kof f s = NOS (kalloc KF ), when
0
Xtotal
0 the subband symbol vector position starts at index kalloc out
of N subcarrier positions. For simplicity of notation, we
have discarded the subband index i so far. This index is
now reintroduced, writing large FFT contribution of the i-th
Fcut(NFFTu-1)
subband as
0
h iT
(i)
0 Xf ull = ~0[1Ko f f s] , XTfilt , ~0[1(NOS N Kof f s NF F T u )] .
0
0 (10)
Fcut(NFFTu)
xtotal
Note that this expression here is written for Kof f s 0. For
Kof f s < 0 the negative spectral contributions have to be
wrapped around in the FFT. The frequency domain result in
Fig. 2. Principle of UF-OFDM frequency domain signal approximation,
illustrating generation of one subband i with NOS = 2. (10) corresponds a time domain output xik of subband i in
Fig. 1. The oversampled frequency domain UF-OFDM signal
is the superposition of all B subbands
B
X
by KF = (NIF F T o Q)/2 zeros on both sides: (i)
Xtotal = Xf ull (11)
h iT
i=1
sext = ~0[1KF ] , sT , ~0[1KF ] (5)
resulting in NOS N time domain samples by the IFFT
The zeros are needed in order to make a certain number of
subband side lobes available in the approximation. A short- xtotal = N IFFT {Xtotal } . (12)
length time domain representation is obtained by carrying out As our UF-OFDM multi-carrier symbol consists of only
an NIF F T o -point IFFT N + L 1 < 2N samples, we cut xtotal to this size, as, due
xshort = IFFT {sext } (6) to the approximated discrete convolution (9), only there the
relevant symbol contributions are contained.
which is expanded by KT = NIF F T o (NOS 1) zeros for
upsampling-purposes, creating a frequency resolution by factor = [xtotal (1), xtotal (2), . . . , xtotal (N + L 1)]T .
x (13)
NOS higher than the subcarrier spacing, by This cut generates a further (small) approximation error.
h iT
xext = xTshort , ~0[1KT ] . (7) III. C OMPLEXITY A SSESSMENT
This is brought into frequency domain by an NF F T u -point In this section we assess the complexity of the multi-carrier
FFT, with NF F T u = NOS NIF F T o , denoted modulation operation. Note that this complexity is only a
Xext = FFT {xext } . (8) fraction of the overall transmitter complexity. Encoder and
RF digital pre-processing consume a lot of operations as well,
Equations (5) - (8) have provided the processing steps for so a 5G system can allow for a modest increase in multi-
upsampling and interpolation. Now the subband FIR-filters are carrier modulation complexity without impacting much the
efficiently carried out in the frequency domain overall transmitter complexity. As a metric for complexity, we
use the total number of real additions and multiplications.
Xf ilt = Xext Fcut (9)
A key operation to be taken into account is the FFT. Each
with Fcut denoting the cut-out frequency response of the complex multiplication is accounted for a total of 6 real
filter, centered around the respective subband position. In operations: 4 real multiplications and 2 real additions. (Note
order to obtain Fcut , we need to start with the time-domain that there exist complex multiplication variants with only 3
low-pass prototype filter coefficients f , designed to reduce real multiplications, with a larger number of additions. Due
the out-of-subband radiation, e.g. according to [10]. As the to a larger total number of operations we have not considered
filter has to be centered onto an even number of subcarri- this 3 multiplication variant in the complexity calculation.)
Complex additions take 2 real additions. We assume the filter
coefficients as precalculated and stored. 0
10 Exact TD
FD N =16
IFFTo
A. FFT- and CP-OFDM-complexity FD NIFFTo=32
1
The divide-and-conquer approach for the FFT by Cooley- 10 FD NIFFTo=64
Tukey is the most commonly used, in conjunction with e.g. FD NIFFTo=128

Amplitude response
a split-radix. The fastest known approach here is [11] which 2
10
forms the basis for our complexity computation: The complex-
ity fF T (N ) of the FFT as a function of its input/output size 3
10
N is
34 124 4
fF T (N ) = N ld N N 2 ld N 10
9 27
2 16
(1)ld N ld N + (1)ld N + 8 (14) 5
10
9 27
50 100 150 200 250 300
where ld represents the binary logarithm. subcarrier index
For CP-OFDM, based on an N -point FFT, the transmitter
complexity for multi-carrier modulation simply is Fig. 3. Spectrum of exact time domain signal generation compared to
approximated frequency domain generation with different NIF F T o ; Q = 12,
cOF DM = fF T (N ). (15) 1 PRB, N = 1024, L = 72, Dolph-Chebychev filters with 40 dB side lobe
attenuation
Any operations regarding CP are neglected here. Hence,
CP-OFDM transmitter complexity order is the well-known
O {N log N }. have to be carried out, as well as B small NOS NIF F T o -sized
FFTs. The frequency domain filtering requires NOS NIF F T o
B. Baseline UF-OFDM complexity complex multiplications per subband.
Let us have a look at the baseline frequency domain All subband contributions have to be added up. Note that
implementation, described in section II.B. For upsampling the we do not account for the additions of known zero positions.
(i)
data symbols in frequency domain according to (2) requires B Thus, the summation in (11) is accounted for as adding Xf ilt
FFTs of size NOS N and B IFFTs of size N . For the frequency (i)
instead of Xf ull . Then a large NOS N -IFFT is made. All this
domain filtering 6BNOS N real operations are needed. Sum- operations for UF-OFDM multicarrier modulation sum up into
ming up the subband contributions leads to 2(B 1)NOS N
real summations. Finally an IFFT of size NOS N is required. cU F,appr = fF T (NOS N )
Thus the overall complexity is + B [fF T (NIF F T o ) + fF T (NOS NIF F T o ) + 6NOS NIF F T o ]
cU F,f ull = B [fF T (NOS N )fF T (N ) + 6NOS N ] + 2(B 1)NOS NIF F T o . (18)
+ 2(B 1)NOS N + fF T (NOS N ) (16) The complexity order for UF-OFDM is thus, like CP-OFDM,
O {N log N }.The overall complexity is at least by the factor
which is at least by factor NOS B higher than CP-OFDM. For
NOS higher than CP-OFDM.
LTE-like multi-carrier parameters with 15 kHz subcarrier spac-
ing and a physical resource block (PRB) size of ni = 12, i, a In the next section absolute complexity numbers are com-
20 MHz bandwidth consists of B = 100 PRBs. So UF-OFDM puted using typical near-LTE multi-carrier system parameters.
in this exact brute-force frequency domain generation case,
IV. N UMERICAL R ESULTS
with full band transmission, would have more than 200 times
the complexity of CP-OFDM. This is not desirable and thus For all the results in this section, the oversampling factor of
an alternative solution has to be designed. the approximation is set to NOS = 2. Our investigations have
For comparison, the matrix-vector-operation-based time- shown that larger numbers NOS > 2 are not required, as they
domain baseline implementation, the textbook approach strongly increase complexity, while satisfying accuracy can
according to (1), for one subband requires also be achieved with NOS = 2. (Further work could consider
FFT-sizes which are not a power of 2. Then NOS < 2 is
cT D,1P RB = (8Q 2)(N + L 1). (17) possible. This allows further potential complexity reductions,
This has to be executed B times. In a full band allocation it making UF-OFDM transmitters almost as simple as CP-
holds BQ N . For Bc OFDM transmitters.)
RB , using (17), the complexity
 T D,1P
order thus becomes O N 2 , as stated in the introduction. Fig. 3 illustrates an example spectrum realization. An UF-
OFDM subband of Q = 12 subcarriers is shown, including
C. Enhanced UF-OFDM complexity the side-lobe levels. NIF F T o = 16 exhibits a very poor side-
For UF-OFDM based on the the frequency domain gen- lobe level behaviour, compared to the exact time domain signal
eration method of this paper, B small NIF F T o -sized IFFTs generation. This is clear as the approximation only uses KF =
0
10 10
overall signal exact TD
15 passband only FD NIFFTo = 16

1
FD N = 32
IFFTo
10
20 FD NIFFTo = 64

MSE after equalization


25
2
10
MSE in dB

30

35
3
10
40

45 4
10

50

5
55 10
0 20 40 60 80 100 120 140 5 10 15 20 25 30 35 40 45 50
N Eb/N0
IFFTo

Fig. 4. Mean squared error at the transmitter output over NIF F T o = Fig. 5. Mean squared error at the equalizer output over Eb /N0 : Exact
{16, 32, 64, 128}, both for overall signal and for pass-band only; Q = 12, 1 time domain signal generation compared to approximated frequency domain
PRB, N = 1024, L = 72. generation with different NIF F T o . Q = 12, 6 PRBs, AWGN channel, 14
symbols per subframe, N = 1024, L = 72, QPSK.

2 subcarriers on both sides of the pass-band to generate the 5


x 10
side-lobe behaviour, which is not enough. NIF F T o = 32 is 7
OFDM
better, but still far from the original. NIF F T o = 64 already Number of real additions and multiplications
6 UFMC TD 1PRB
achieves the envelope level of the side-lobes and thus can be UFMC FD 1PRB
considered as satisfying. NIF F T o = 128 is even better. UFMC FD 5PRB
5 UFMC FD 20 PRB
Fig. 4 depicts the MSE at the transmitter output 
for one sub-
UFMC FD 50 PRB
band. For the overall signal, we define this as E |x x |2 , 4
the distortions to the exactly generated vector x. For the pass-
band, we measure the frequency domain distortions caused 3
by Xtotal at the data-carrying subband positions. We can
observe that the pass-band of one subband is already perfectly 2
approximated with NIF F T o 32, as the MSE is 50 dB.
For the overall signal the difference is larger, but signal 1

distortions occurring in the side-lobe parts of the spectrum


0
are not as important as in the pass-band, as long as the 0 20 40 60 80 100 120 140
Small FFT size
overall side-lobe level and thus out-of-band-radiation remains
approximately unchanged (see Fig. 3). We can consider UF-
OFDM as adequatly accurate, when the signal distortions of Fig. 6. Complexity as a function of NIF F T o for N = 1024, NOS = 2:
Number of real additions and multiplications.
the approximation method are below distortions from the RF
chain impairments. For 64-QAM, 3GPP tolerates an EVM of
8% [12], which corresponds to an MSE of 22 dB. This time-domain UF-OFDM signal generation (1) for an allocation
is clearly fulfilled for NIF F T o 64. of one PRB has about factor 3 higher complexity, using (17).
Fig. 5 depicts the MSE at the equalizer output. QPSK sym- It scales linear with the number of PRBs B. In 10 MHz
bols of 6 subbands are mapped onto UF-OFDM subcarriers with LTE-like parameters, there are 600 allocated subcarriers,
and transmitted via an AWGN channel. The receiver collects thus 50 PRBs. This brute force textbook approach in time
the N + L 1 samples of a multi-carrier symbol, see e.g. domain would result in UF-OFDM multi-carrier modulator
[13]. The symbol is zero-padded up to a length of 2N . At the complexity of about 150 times CP-OFDM complexity. For the
output of a 2N -point FFT, each second value represents the approximative frequency domain generation, 1 PRB consumes
subcarrier symbol. A scalar single-tap equalizer simply divides about factor of 2.4 more than CP-OFDM. While this value is
by the estimated channel and the known frequency domain growing with the number of allocated subbands, the slope of
filter coefficient in order to get the equalized
n P symbol vector ori the growth is much smaller than in the exact method, as the
B
for subband i. The MSE is defined as E B1 i=1 |si ri |2 . signal is constructed by those small IFFTs and FFTs instead
Fig. 6 shows the complexity in terms of real multiplications of full size vectors.
and additions, based on the computations of section III. CP- Fig. 7 shows the complexity normalized to CP-OFDM
OFDM complexity is independent of allocation size. Baseline complexity. The larger the FFT size, the smaller becomes
R EFERENCES
11
N=512 [1] 5GNOW, http://www.5gnow.eu/
10 N=1024 [2] METIS, https://www.metis2020.com/
real add. and multi. normalized over OFDM

N=2048 [3] Wunder, G.; Jung, P.; Kasparick, M.; Wild, T.; Schaich, F.; Chen, Y.;
9 Brink, S.T. et. al. 5GNOW: non-orthogonal, asynchronous waveforms
for future mobile applications, Communications Magazine, IEEE, vol.52,
8 no.2, pp.97,105, Feb. 2014
[4] Farhang-Boroujeny, B.; OFDM versus Filter Bank Multicarrier, IEEE
7 Signal Process. Mag., Vol. 28, pp. 92112, May 2011
[5] Fettweis, G.; Krondorf, M.; Bittner, S., GFDM - Generalized Frequency
6 Division Multiplexing, IEEE Vehicular Technology Conference, VTC
Spring 2009, IEEE 69th, pp.1-4, 26-29 April 2009.
5
[6] Schaich, F.; Wild, T.; Chen, Y.; Waveform contenders for 5G - suitability
for short packet and low latency transmissions, in proc. IEEE Veh.
4
Technol. Conf. Spring (VTC14 Spring), May 2014.
[7] Wild, T.; Schaich, F.; Chen, Y.; 5G Air Interface Design based on
3
Universal Filtered (UF-)OFDM, in proc. DSP14, Hongkong, Aug. 2014.
2
[8] Schaich, F.; Wild, T.; Relaxed Synchronization Support of Universal
0 20 40 60 80 100 Filtered Multi-Carrier including Autonomous Timing Advance, in proc.
number of subbands (PRBs) IEEE ISWCS14, Barcelona, Aug. 2014.
[9] M. Bellanger et al., FBMC physical layer : a primer, http://www.ict-
phydyas.org/
Fig. 7. UF-OFDM complexity with NIF F T o = 64, NOS = 2 for different
[10] Wang, X.; Wild, T.; Schaich, F.; Santos, A., Universal Filtered Multi-
FFT sizes normalized to CP-OFDM.
Carrier with Leakage-Based Filter Optimization, European Wireless,
EW14, Barcelona, May 2014.
[11] S. G. Johnson and M. Frigo, 2007. A modified split-radix FFT with
the complexity growth as a function of B. This is because fewer arithmetic operations, IEEE Trans. Signal Processing 55 (1):
the large FFT size is dominating more over the small FFT 111119.
[12] 3GPP TS 36.104, 3rd Generation Partnership Project; Technical Specifi-
sizes NIF F T o . Allocating one PRB costs factor 2.4 to 2.6 cation Group RAN4; E-UTRA Base Station (BS) radio transmission and
more complexity than CP-OFDM. Allocating an entire band reception (Release 11), V11.11.0.
roughly requires between factor 8 and factor 10 more mul- [13] Chen, Y.; Schaich, F.; Wild, T.; Multiple Access and Waveforms for
5G: IDMA and Universal Filtered Multi-Carrier, in proc. IEEE VTCs14,
tiplications and additions than CP-OFDM. For example a Seoul, Korea, May 2014.
2048-FFT corresponds to 20 MHz bandwidth with LTE multi-
carrier numerology. For this 2048-FFT, the complexity of the
UF-OFDM frequency domain approximation method is by
more than factor 30 reduced, compared to the exact baseline
implementation.
V. C ONCLUSION
UF-OFDM is a promising 5G waveform candidate technol-
ogy. We have provided a UF-OFDM multi-carrier modulator
for enabling an efficient hardware implementation. It is fre-
quency domain based and uses approximations for complexity
savings. The solution uses adjustable parameters NIF F T o .
Here reasonable values of e.g. NIF F T o = 64 lead to negligible
approximation errors, as proved by assessing the mean squared
error and the spectral properties. With this approach, the UF-
OFDM complexity order is identical to CP-OFDM. The UF-
OFDM overall complexity in terms of multiplications and
additions is between factor of 2.4 and 10 higher than CP-
OFDM, when an oversampling factor NOS = 2 is used.
Our novel approach is more than one order of magnitude
less complex than so-far existing transmitters for UF-OFDM.
For those reasons, this presented approach will be included
in upcoming hardware demonstrators for UF-OFDM. Future
work should consider FFT-sizes which are not a power of 2,
checking for further potential complexity reductions.

Das könnte Ihnen auch gefallen