Professional Documents
Culture Documents
on each subcarriers. By modulating these symbols with the Based on these provided coefficients, the frequency response
inverse fast Fourier transform (IFFT) at the transmitting side, of the filter with M subcarriers is obtained through the
a discrete baseband OFDM symbol can be generated, which equation
can be written as
k
1
M −1 K−1
sin π f − M K
x[n] = F −1 {X[i]} = √
2π
X[i]ej M ni , 0 ≤ n ≤ M −1, MK
P (f ) = Pk . (4)
M i=0 k
(1) k=−(K−1) M K sin π f −
MK
where, M is the number of subcarriers, X[i] is a complex
QAM symbol transmitted on the ith subcarrier. As shown in Figure 4(a), a highly frequency selective filter
Obviously, the QAM symbol can be obtained at the receiver is acquired with almost no out of band leakage. Therefore,
side utilizing Fast Fourier transform (FFT). In the perfect
reconstruction situation, the recovered signal is
M −1
1 2π
X̂[i] = F{x[n]} = √ x[n]e−j M ni , 0 ≤ i ≤ M − 1.
M n=0
(2)
QAM OQAM
T T
2 2
n−2 n−2
n−1 n−1
(a) Prototype filter frequency response (b) Prototype filter impulse response
Fig. 4. Prototype filter [4]
n n
the impulse response p[m] of the prototype filter is given by
n+1 n+1 applying the IFFT to the frequency response, which is
K−1
n+2 n+2 k 2πk
p[m] = 1 + 2 (−1) Pk cos m
MK (5)
Real part of QAM symbol Imaginary part of QAM symbol k=1
d1(mM ) I H1 Z −1
F
x(n)
F
T
Fig. 5. Filter bank generated from prototype filter [6]
dM −1(mM ) HM −1 Z −(M −1) With this prototype filter, a filter bank of the system is ob-
tained with frequency shifts, as shown in Figure 5. Obviously,
Fig. 3. PPN-IFFT FBMC transmitter [4] the subcarriers with odd (or even) index are not overlapped.
In this work, the PPN filter applied is from PHYDYAS [4] Only the neighbour subcarriers have influences in a certain
with the overlapping factor K = 4. The frequency domain subcarrier. On account of this feature, one can exploit only odd
coefficients of it are or even subcarriers for communication with QAM symbols.
√
2 However, by utilizing only alternate subcarriers, the bit rate
P0 = 1; P±1 = 0.97196; P±2 = 2 ; P±3 = 0.235147.
comes down by half as half of the capacity is left unused.
And they satisfy the equation To gain the full capacity, orthogonality is necessary for the
K−1 neighbouring sub-channels. This orthogonality is satisfied with
1
|Pk |2 = 1. (3) offset quadrature amplitude modulation (OQAM). The OQAM
K
k=−K+1
p[0] p[1] p[2] p[3] p[4] p[5] p[6] p[7] p[8] p[9] p[10] p[11] p[12] p[13] p[14] p[15]
y1 [n] = p[n]x1 [n%M ]
x1 [0] x1 [1] x1 [2] x1 [3] x1 [0] x1 [1] x1 [2] x1 [3] x1 [0] x1 [1] x1 [2] x1 [3] x1 [0] x1 [1] x1 [2] x1 [3]
p[0] p[1] p[2] p[3] p[4] p[5] p[6] p[7] p[8] p[9] p[10] p[11] p[12] p[13] p[14] p[15]
y2 [n] = p[n]x2 [n%M ]
x2 [0] x2 [1] x2 [2] x2 [3] x2 [0] x2 [1] x2 [2] x2 [3] x2 [0] x2 [1] x2 [2] x2 [3] x2 [0] x2 [1] x2 [2] x2 [3]
p[0] p[1] p[2] p[3] p[4] p[5] p[6] p[7] p[8] p[9] p[10] p[11] p[12] p[13] p[14] p[15]
x3 [0] x3 [1] x3 [2] x3 [3] x3 [0] x3 [1] x3 [2] x3 [3] x3 [0] x3 [1] x3 [2] x3 [3] x3 [0] x3 [1] x3 [2] x3 [3]
p[0] p[1] p[2] p[3] p[4] p[5] p[6] p[7] p[8] p[9] p[10] p[11] p[12] p[13] p[14] p[15]
x4 [0] x4 [1] x4 [2] x4 [3] x4 [0] x4 [1] x4 [2] x4 [3] x4 [0] x4 [1] x4 [2] x4 [3] x4 [0] x4 [1] x4 [2] x4 [3]
y[0] y[1] y[2] y[3] y[4] y[5] y[6] y[7] y[8] y[9] y[10] y[11] y[12] y[13] y[14] y[15] y[16] y[17] y[18] y[19] y[20] y[21]
staggered the in-phase and quadrature components by half a 1) Divide the signal into K frames and apply the prototype
symbol period. Figure 2 implies the difference between QAM filter to each frame. It is easy to see that for ρth frame we
and OQAM symbols intuitively. In each sub-channel either the have rρ [n] = r[n + ρ−1
2 M ]. Therefore, by multiplying the ρth
real or the imaginary part is transmitted avoiding interference frame with the impulse response of the prototype filter, we
between neighbouring sub-channels. Finally in order to keep have
up the bit rate, the FFT is run at twice the rate as in OFDM. sρ [n] = p[n]rρ [n]. (9)
Now, we consider the signal at the transmitter side of
the FBMC. As show in Figure 1, after the serial-to-parallel 2) Resize the signal from the last step and sum up. Recall
conversion, the OQAM symbols are passed into the IFFT that the symbol after IFFT is duplicated K times at the
block. Be denoting Xρ [i] as the symbol on the ith carrier of transmitter side. Therefore, these K symbols must be reformed
ρth frame, we obtain the IFFT block output xρ [n], and summed up back to one symbol again. We use ŝρ [n] to
M −1 denote this symbol, then
1
xρ [n] = F −1 {Xρ [i]} = √ Xρ [i]ej2πni/M . (6) K−1
M i=0
ŝρ [n] = sρ [n + kM ] n = 0, 1, ..., M − 1. (10)
These symbols will be treated with the prototype filter before k=0
transmitting. Figure 6 shows how this symbol is constructed,
in which both the overlapping factor and the carrier size are At this point, we can pass the symbol ŝρ [n] to the FFT
equal to 4. Obviously, the procedure to form such a symbol block frame by frame to recreate the OQAM symbol, which
can be accomplished by two steps: is
1) Treat each frame with the prototype filter. As we can M −1
see from Figure 6, the output of IFFT xρ [n] is duplicated K 1
X̂ρ [i] = F{ŝρ [n]} = √ ŝρ [n]e−j2πni/M . (11)
times and then multiplied with the impulse response of the M n=0
prototype filter. In order to simplify this procedure,
we first
define the modulo operation as a%b = a − ab × b, where III. S YSTEMS UNDER AWGN CHANNEL
a, b are integers and x means the largest integer not greater
than x. Now, each filtered frame symbols yρ,n is In this section, we compare the performance of these two
systems under the AWGN channel and acquire the analytical
yρ [n] = p[n]xρ [n%M ] . (7) BEP expressions for them simultaneously. The Gaussian noise
is defined as w ∼ N (0, N0 ).
2) Overlap the frames with half symbol duration shift. From
the previous step, we get the filtered frames. Now they must
be shifted with half symbol period M/2 one by one. Thus, A. OFDM based model
the final transmitting signal is obtained by In the AWGN channel, the CP component of the OFDM
K system can be omitted as there is no ISI occurred. Hence,
ρ−1 by denoting the received signal as r[n] = x[n] + w[n] and
y[n] = yρ [n − M] . (8)
ρ=1
2 combing equation (1) with equation (2) , we have
Because of the shift, the total length of yn will be extended, X̂[i] = F{r[n]} = F{x[n] + w[n]}
which is equal to Lp + (K − 1)M/2 = (3K − 1)M/2. = F{x[n]} + F{w[n]}
At the receiver side of the FBMC, a two steps process is (12)
= F{F −1 {X[i]}} + F{w[n]}
employed to recover the signal which will be passed into the
FFT block, as shown in Figure 7. Let r[n] denotes the received = X[i] + W [i].
signal, then:
y2 [n]
y1 [n]
y[0] y[1] y[2] y[3] y[4] y[5] y[6] y[7] y[8] y[9] y[10] y[11] y[12] y[13] y[14] y[15] y[16] y[17] y[18] y[19] y[20] y[21]
p[0] p[1] p[2] p[3] p[4] p[5] p[6] p[7] p[8] p[9] p[10] p[11] p[12] p[13] p[14] p[15]
s1 [0] s1 [1] s1 [2] s1 [3] s1 [4] s1 [5] s1 [6] s1 [7] s1 [8] s1 [9] s1 [10] s1 [11] s1 [12] s1 [13] s1 [14] s1 [15]
s1 [n] s1 [n + M ] s1 [n + 2M ] s1 [n + 3M ]
ρ−1
From equation (12) , we can see that W [i] is the noise term in which r[n + kM + 2 M] can be simplified as
which is corrupting the QAM signal. Thus, the power of it ρ−1
should be obtained for the BER expression, which is r[n + kM + M]
2
K
⎧
ρ−1 ρ−1
2 ⎫ = p[n + kM + M− M ]·
⎨
1
N
−1 ⎬ 2 2
2π ρ=1
E W 2 [i] = E √ w[n]e−j N ni
⎩ N n=0 ⎭ ρ−1 ρ−1
(13) xρ [ n + kM + M− M %M ] (16)
N −1 2 2
1
=
4π
E w2 [n] e−j N ni = N0 . ρ−1
N + w[n + kM + M]
n=0 2
K
= p[n + kM ]xρ,n%M + wn+kM + ρ−1 M .
2
B. FBMC based model ρ=1
Where, F{p̂2 [n]xρ [n]} can be expressed as C. BER derivation for both systems
From the two previous subsectoins, the power of the noise
F{p̂2 [n]xρ [n]} that influences QAM and OQAM demodulators are derived,
M −1
1 2 both are N0 . And it is clear that OQAM can be treated as a
=√ p̂ [n]F −1 {Xρ [i]}e−j2πni/M special case QAM because it only separates a QAM symbol
M n=0 into two symbols. This means that for I-ary rectangular QAM
M −1 M −1 or OQAM symbol, the BER expression is the same under the
1 2 1
=√ p̂ [n] √ Xρ [i]ej2πni/M e−j2πni/M AWGN channel for both OFDM and FBMC, which is
M n=0 M i=0 √
M −1 I −1 3 log2 I Eb
1 2 (19) Pb ≈ √ √ erfc ·
2 (I − 1) N0
, see [7]. (24)
= Xρ [i] p̂ [n] I log2 I
M n=0
M −1 K−1
IV. S YSTEMS UNDER R AYLEIGH C HANNEL
1 2 In this section, the two systems under Rayleigh channel are
= Xρ [i] p [n + kM ]
M n=0 studied. We use h[n] to denote the impulse response of the
k=0
Lp−1
Channel.
1
= Xρ [i] p2 [n]. A. OFDM
M n=0 In this case, the received signal is r[n] = h[n] ∗ x[n] + w[n],
1
Lp−1 2 where ∗ meas convolution. At the receiver side for OFDM, a
As in equation (19), the term M n=0 p [n] = 1. Thus, we frequency domain equalizer will be applied to mitigate the
can recover the signal with the added noise, which is, distortion introduced by the channel, denoted as C[i]. The
equalizer will satisfy the constraint which is
X̂ρ [i] = Xρ [i] + F{ŵρ [n]} = Xρ [i] + Ŵρ [i].
(20) 1
C[i] = , (25)
H[i]
The equation above shows that the noise term at the OQAM where H[i] is the frequency response of the channel. There-
demodulator is Ŵρ [i]. As a result, the power of it should be fore, the signal entering the QAM demodulator is
obtained to get the analytical BER expression.
X̂[i] = C[i] · F{r[n]} = C[i] · F{h[n] ∗ x[n] + w[n]}
E Ŵρ2 [i] = E F 2 {ŵρ [n]} = C[i]F{h[n] ∗ x[n]} + C[i]F{w[n]}
⎧
2 ⎫
(26)
⎨ M −1 ⎬ = C[i]H[i]F{F −1 {X[i]}} + C[i]F{w[n]}
1
=E √ ŵρ [n]e−j2πni/M = X[i] + C[i]W [i] = X[i] + Ŵ [i].
⎩ M ⎭
n=0
Thus, the channel attenuated noise power is
M −1 (21)
1 2 −j4πni/M 2
= E ŵρ [n] e E Ŵ 2 [i] = E (C[i]W [i])
M n=0
M −1 = E C 2 [i] · E W 2 [i] (27)
1 2
= E ŵρ [n] . = N0 E C 2 [i] .
M n=0
As a result, the average SNR per bit is
It can be derived as Eb Eb 2 Eb
γ= = E H [i] = Ω. (28)
E ŵρ2 [n] N0 E {C 2 [i]} N0 N0
⎧
2 ⎫
⎨ K−1 ⎬ B. FBMC
ρ−1
=E p[n + kM ]w[n + kM + M] For the PPN based FBMC system, a time domain equalizer
⎩ 2 ⎭ will be applied for the compensation of channel fading. By
k=0
K−1 denoting the impulse response of the equalizer as c[n] with
ρ−1 property of
= p2 [n + kM ]E w2 [n + kM + M]
2 c[n] ∗ h[n] = δ[n], (29)
k=0
K−1
where δ[n] is the Dirac delta function, we can achieve the
= N0 p2 [n + kM ]. received signal as rn = h[n] ∗ y[n] + w[n]. Thus, by passing
k=0 it through the equalizer, the mitigated sequence r̂[n] is
(22)
r̂[n] = c[n] ∗ r[n] = c[n] ∗ (h[n] ∗ y[n] + w[n])
By substituting Equation 22 into Equation 21, we get, = y[n] + c[n] ∗ w[n] . (30)
M −1 K−1 v[n]
1
E Ŵρ [i] = N0 p2 [n + kM ] = N0 . (23) As we define c[n] ∗ w[n] = v[n], then r̂[n] has the same
M n=0
k=0 structure as equation (14) in Section III, which means the
same derivations are obtained here. Using equation (20), the V. S IMULATION AND EVALUATION
received signal of the OQAM demodulator are,
In this section, the theoretical and simulated BER will
be compared under both AWGN and Rayleigh channels for
X̂ρ [i] = Xρ [i] + F{v̂ρ [n]} = Xρ [i] + V̂ρ [i]. (31) FBMC and OFDM. The carrier size M of them are 64. And
the 64-QAM (or OQAM) is applied in the simulation. For the
Now, we consider V̂ρ,i , which is OFDM system, the length of CP is 16. And in the FBMC
system, the overlapping factor K = 4 is used. The Rayleigh
V̂ρ [i] = F{v̂ρ [n]} multipath channel applied here is a 5-tap channel. We assume
K−1 that the perfect channel state information is available at the
ρ−1 receiver.
=F p[n + kM ]v[n + kM + M]
2
k=0
M −1 K−1 (32) A. Under AWGN Channel
1
=√ p[n + kM ]· For the AWGN channel, the CP part of OFDM system can
M n=0 k=0 be neglected as there is no ISI. Therefore, the CP are not taken
ρ−1 into the simulation.
v[n + kM + M ]e−j2πni/M .
2 0
10
OFDMSimulation
Similar as equation (22), we get Theoretical
1
10
2 2 ρ−1
E V̂ρ [i] = E v [n + kM + M]
2 2
10
2
= E v 2 [λ] = E (c[λ] ∗ w[λ])
Bit Error Rate
⎧
2 ⎫ 3
10
⎨ ⎬
=E w[η]c[λ − η]
⎩ η ⎭ 4
10
= E (w[η]c[λ − η])
2 (33)
5
η 10
= E w2 [η] E c2 [λ − η]
6
η 10
0 2 4 6 8 10 12 14 16 18
= N0 E c2 [λ − η] EbN0 (dB)
1
10
6taps
60taps 2
10
200taps 0 2 4 6 8 10 12 14 16 18 20
600taps EbN0 (dB)
1200taps Fig. 11. OFDM without CP in the Rayleigh channel
Theoretical
2
10
0 2 4 6 8 10 12 14 16 18 20
EbN0 (dB) VI. CONCLUSION
Fig. 9. FBMC with 200-tap equalizer in the Rayleigh Channel In this paper, the analytical BER expressions for OFDM and
PPN-FBMC systems under AWGN and Rayleigh channels are
From the figure above, we can tell that in order to reach derived and validated. We prove that the BER performance of
the theoretical BER performance for FBMC system, a very the two systems are the same when the perfect recover condi-
long time domain equalizer is necessary. However, this will tions are satisfied. In order to achieve this, OFDM requires the
introduce extra time delay to the system with growth in the cyclic prefix to eliminate the inter symbol interference, while
computational complexity. FBMC needs a long time domain equalizer.
Secondly, the OFDM system is simulated and evaluated. Hence, the further research can concentrate on the new
Figure 10 and Figure 11 show the BER performance of this equalization technique for PPN-FBMC which reduces the
system in the situation with and without CP correspondingly. complexity for mitigating the signal.
0
10
OFDMSimulation
Theoretical
R EFERENCES
[1] A. Sahin, I. Guvenc, and H. Arslan, “A survey on multicarrier communi-
cations: Prototype filters, lattice structures, and implementation aspects,”
Communications Surveys Tutorials, IEEE, vol. 16, no. 3, pp. 1312–1338,
Third 2014.
[2] Q. He, C. Schmitz, and A. Schmeink, “BEP of fourier transform and
Bit Error Rate