Вы находитесь на странице: 1из 5

Performance Evaluation of SC-FDMA Uplink Block Transmission

Schemes Using Iterative Receiver Techniques


L. Charrua*, P. Torres**, V. Gonçalves***, A. Gusmão****
CAPS-IST, Technical University of Lisbon, Portugal, Email: luischarrua@nautica.pt*, Email:
paulo.torres@ieee.org**, Email: victor.goncalves@nautica.pt*** Email: gus@ist.utl.pt****

Abstract—This paper deals with CP-assisted block transmis- users, within the set of subcarriers provided by the system.
sion solutions for future mobile broadband systems, in the context OFDMA (Orthogonal Frequency Division Multiple Access)
of a SC (Single Carrier)-FDMA (Frequency-Division Multiple is a recommended choice for downlink transmission, where
Access) uplink. Two alternative choices are considered regarding
the subcarrier mapping rule: a ”localized” subcarrier mapping frequency-domain data symbols (e.g. from a QAM (Quadra-
where user’s data occupy a set of consecutive sub-carriers (Rule ture Amplitude Modulation) or PSK (Phase Shift Keying)
1); a ”distributed” subcarrier mapping where user’s data occupy alphabet) are used; SC-FDMA (Frequency-Division Multiple
a set of uniformly spaced subcarriers (Rule2). Access) is a recommended choice for uplink transmission,
Detailed performance evaluations, in this paper, involve the where time-domain data symbols (from the same QAM or
consideration of two iterative receiver techniques which can be
regarded as extensions of iterative receiver techniques proposed PSK alphabets) are adopted. It should be noted that current
previously within a single user context. A selected class of working assumptions for the air interface in the 3rd Generation
multipath radio channels, providing a range of channel time Partnership Project/Long Term Evolution (3GPP LTE) follow
dispersion levels, is assumed for performance evaluation pur- this hybrid approach [???].
poses, and a set of matched filter bounds on receiver performance In this paper, we consider SC-FDMA uplink block trans-
plays a relevant role in ”achievable performance” comparisons.
Both the impact of the mapping rules and that of the iterative mission using two alternative choices in which concerns the
receiver techniques considered here are evaluated in detail. The subcarrier mapping rule: a ”localized” subcarrier mapping
performance advantages under ”Rule 2” are emphasized, for the where use’s data occupy a set of consecutive, adjacent sub-
entire range of channel dispersion levels assumed in the paper carriers (Rule 1); a distributed subcarrier mapping where user’s
and both specific iterative receiver techniques. data occupy a set of uniformly spaced subcarriers (Rule 2). The
latter mapping rule, in accordance with a proposal in [????]
(later adopted by other authors, under several acronyms), is
I. I NTRODUCTION
known to provide a lower envelope fluctuation. The goal of
In recent years, appropriately designed CP (Cyclic Prefix)- this paper is to evaluate, in detail, the receiver performance
assisted block transmission schemes were proposed and devel- advantage under this rule. Performance evaluations, in this
oped for broadband wireless systems, which have to deal with paper, involve the consideration of iterative receiver techniques
strongly frequency-selective fading channel conditions. These with two levels of complexity, which can be regarded as
schemes take advantage of current low-cost, flexible, FFT an extension of the iterative receiver approach proposed in
(Fast Fourier Transform)- based signal processing technology [GTDE06], [GTDE07] regarding a single-user context. A
with both OFDM (Orthogonal Frequency Division Multi- selected class of multipath radio channels, providing a range of
plexing) and SC (Single Carrier)/FDE (Frequency Domain channel time dispersion levels, is considered for performance
Equalization) alternative choices [SKJ94]. Mixed air interface evaluation purposes, and a set of matched filter bounds on
solutions, with OFDM for the downlink and SC/FDE for receiver performance plays a relevant role in ”achievable
the uplink as proposed in [GDCE00], [FABSE02] are now performance” comparisons. Both the impact of the mapping
widely accepted. The main reason for replacing OFDM by rules and that of the iterative receiver techniques considered
SC/FDE, with regard to uplink transmission, is the lower here are evaluated in detail.
envelope fluctuation of the transmitted signals when data
symbols are directly defined in the time domain, leading to
reduced power amplification problems at the mobile terminals. II. SC-FDMA T RANSMISSION M ODELS AND U NCODED
”Clipped and filtered” OFDM schemes have also been consid- P ERFORMANCE B OUNDS
ered for low-PMEPR (Peak-to-Mean Envelope Power Ratio)
A. Transmission Assumptions
uplink transmission, in recent years. However, the SC/FDE
alternative was shown to provide clearly better overall power rrr
efficiency when considering advanced iterative techniques of The main transmission assumptions are stated in the fol-
similar complexity in both cases [TG07]. Regarding broad- lowing, with the help of the simplified uplink transmitter
band wireless communication systems based on CP-assisted structure depicted in Fig. 1. Under the SC-FDMA assumption,
block transmission, also became widely accepted, in recent the orthogonality of simultaneous users comes from the fact
years, multiple access techniques based on the assignment that each user occupies different subcarriers. With J users
of disjoint subsets of subcarriers to the several simultaneous and N subcarriers provided by the system, this implies that

1
PJ
j=1 Mj ≤ N , when using Mj to denote the number of sub- B. Channel Model
carriers occupied by user j. The length-Mj block of time- In this paper, we consider a class of multipath radio chan-
domain input data symbols concerning user j is given by nels, all of them characterized by a given set of path ”power
(in)
h
(in) (in) (in)
iT gains” {Pi ; i = 0, 1, · · · , 10} and a corresponding set of path
sj = sj,0 , sj,1 , · · · , sj,M −1 (1)
delays {τi ; i = 0, 1, · · · , 10}. We assume that τi = iτ with
where we assume, throughout this paper, QPSK (Quater- τ = Ns Ts , where Ts is the symbol duration defined in subsec.
(in) II-A and Ns is an integer that can be adjusted according to
nary Phase  symbols, i.e. sj,n = (±1 ±
 Shift Keying)
σ (in) 2 the intended level of time dispersion effects on the transmitted
j) √j2 E sj,n = σj2 . This block results from the block signals. We also assume exponentially decreasing values for
of data bits through channel coding, interleaving, modulation the Pi coefficients, as shown in Fig. 2, and an independent
and serial-to-parallel operations. Through an Mj -point DFT Rayleigh fading for each path and each receiver branch.
(Discrete Fourier Transform), a frequency-domain representa-
tion of the data block is then produced, according to
h iT
(in) (in) (in) (in) (in)
Sj = FMj sj = Sj,0 , Sj,1 , · · · , Sj,M −1 , (2)
and this is followed by a selected procedure, where each of
the Mj DFT outputs is mapped into one of the N (≥ Mj )
orthogonal subcarriers that can be transmitted. The result of
this subcarrier mapping is the length-N frequency-domain
vector
T
Sj = [Sj,0 , Sj,1 , · · · , Sj,N −1 ] (3)
Fig. 2. Example of ”power profile” {Pi ; i = 0, 1, · · · , 10}.

For the jth user and the qth receiver branch at the
base station (q = 1, , 2, · · · , Q), the CIR (Channel Impulse
h iT
(q) (q) (q) (q)
Response) hj = hj,0 , hj,1 , · · · , hj,N −1 is then char-
acterized as follows, through the use of independent, zero-
(q)
mean, complex Gaussian variables Zj,i with variance equal
(q) p (q)
a 1: hj,n = Pn/Ns Zj,n/Ns , if n mod Ns = 0, and zero
h i
(q)
otherwise (n = 0, 1, · · · , N − 1). Of course, E hj,n = 0,
 
(q) 2
E hj,n = Pn/Ns if n mod Ns = 0 and zero otherwise,
(q) (q 0 )∗
h i
and E hj,n hj 0 ,n0 = 0 for n0 6= n, q 0 6= q and/or j 0 6= j.
Fig. 1. Uplink transmitter structure. The CFR (Channel Frequency h Response) correspondingi T
(q) (q) (q) (q) (q) (q)
to hj is Hj = FN hj = Hj,0 , Hj,1 , · · · , Hj,N −1 .
Then, an IDFT (Inverse Discrete Fourier Transform) opera-
When the CP is long enough to cope with the channel time
tion brings the data block information back to the time domain,
dispersion, the frequency-domain received samples concerning
leading to the length-N vector
user j (after removal of the LS CP-related samples and a
sj = F−1
N Sj , (4) length-N DFT), at the qth branch of the BS (Base Station)
and, by inserting a length-Ls CP, the transmitted symbol block receiver, are given by
is obtained. By assuming that the time-domain transmitted (q) (q) (q)
Yj,k = Sj,k Hj,k + Nk , (7)
symbols sj,n (components of sj ) have duration Ts , the CP
(q)
duration and the duration of the transmitted block can be where Nk corresponds to independent, zero-mean,
expressed as Ls Ts and (N + Ls )Ts , respectively. frequency-domain Gaussian noise terms with variance
Two subcarrier mapping rules considered here are described N N0 . Of course, this happens for LS ≥ 10NS . Otherwise,
in the following. (q) (q) (q) (q)
X (q)
Yj,k = Sj,k Hj,k + Nk + Xj,k + Xj 0 ,k (8)
Mapping rule R1 : for a selected Kj ,
j 0 6=j
(in)
Sj,k = Sj,k−Kj (5) for each sub-channel k concerning user j, where the last
if k = Kj , Kj + 1, · · · , Kj + Mj − 1. two terms at the right-hand side of (8) correspond to self-
Mapping rule R2 : for a selected Kj0 [0, mj − 1], interference and multi-user interference resulting from the
insufficient-CP conditions (LS < 10NS , i.e. a CP duration
(in)
Sj,k = Sj,(k−K 0 )/mj (6) LS TS below 10τ = 10NS TS ).
j
  For user j and branch q, we get a length-
if (k − Kj0 )mod mj = 0 mj = M N
j
. Mj , frequency-domain, input vector Yj
(in),(q)
=

2
bounds can be obtained through a simple and fast computation,
under the channel assumptions of sec II-B. This is the case
with the MFB (Matched Filter Bound) concerning ”raw” BER
(Bit Error Rate) performance, i.e., the error rate for direct
decisions on the coded bits, without decoding. Under the
ideal ”ISI (Inter-Symbol Interference)-free” assumption of a
single user (j) transmission with a single symbol per block,
(in) (in) σ
we get an input vector sj with sj,n = √j2 (±1 ± j)
0
for n = n ∈ (0, Mj − 1) and zero otherwise, leading
(in) (in)
to Sj,k = sj,n0 exp −j2πn0 k/Mj (k = 0, 1, · · · , Mj − 1).
Therefore, for a CP long enough to cope with the channel
(q),(in)
(A) time dispersion, the received Yj vector has components
0
 
(q),(in) (in) −j2πn k (q),(in) (q),(in)
Yj,k = sj,n0 exp Hj,k + Nj,k (10)
Mj
(q = 1, · · · , Q; k = 0, 1, · · · , Mj − 1). Having in mind the
Q-branch, FDE scheme of Fig. 3 B, it is easy to obtain
(in) (in)
s̃j,n0 = γj sj,n0 + (11)
Mj −1 Q
j2πn0 k
 X
1 X (q),(in)∗ (q),(in)
Cj,k exp Hj,k Nj,k
Mj Mj q=1
k=0

(since Cj = [Cj,0 , Cj,1 , · · · , Cj,Mj −1 ]), with


(B) Mj −1 Q
1 X (q),(in) 2
X
Fig. 3. Generation of the user input vectors, at the BS receiver, concerning γj = Cj,k Hj,k (12)
Mj

the several users (A) and subsequent FDE for user j (B). k=0 q=1

as an ”equivalent channel” gain.


h
(in),(q) (in),(q) (in),(q)
iT Matched-filter conditions correspond to Cj,k =
Yj,0 , Yj,1 , · · · , Yj,Mj −1 which is related C (constant). For a given ”channel realization” (a given
(q) (q),(in)
to the set of Mj components of the length-N Yj,k set {Hj ; q = 1, · · · , Q}), and taking into a account
vector Y(q) concerning user j. Under the use of rules the variance N N0 of each Nj,k
(q),(in)
in (11), the conditional
R1 and R2 , respectively, we can write the following:

(q),(in)
(q) (q),(in) (q) (q),(in) (q) (q),(in) ”raw” BER is easily derived as P (e {Hj ; q =
Yj,k = Yj,k−Kj , Hj,k = Hj,k−Kj , and Nk = Nj,k−Kj , if r !
σj2 PMj −1 PQ (q),(in) 2
(q) (q),(out)

k = Kj , Kj + 1, · · · , Kj + Mj − 1; and Yj,k = Yj,(k−K 0 )/mj , 1, 2, · · · , Q}) = Q k=0 q=1 Hj,k ,
j N0 N
(q) (q),(in) (q) (q),(in)
Hj,k = Hj,(k−K 0 )/mj , and Nk = Nj,(k−K 0 )/mj , if where Q(.) denotes the well-known Gaussian error function.
j j
(k − Kj0 ) mod mj = 0. Of course, eqn. (7) can be rewritten Having in mind  that (by averaging over all channel
 P
as (q),(in) 2 N −1
(q),(in) (in) (q),(in) (q),(in) realizations), E Hj,k = n=0 Pn , according to the
Yj,k = Sj,k Hj,k + Nj,k (9)
Eb
channel model of sec. II-B, we can express the average N
(k = 0, 1, · · · , Mj − 1) and similarly for eqn. (8). The signal 0
per receiver branch (with Eb denoting the average coded
processing operations described above (CP removal, length-N Eb Mj σ 2 P
N −1
DFT and user ”separation”) are performed in the BS receiver, bit energy) as N 0
= 2ηNj n=0 Pn where η = N N ∗Ls .
as depicted in Fig. 3.A. This user separation involves the ap- Therefore, the average BER can be easily, semi-analytically
Eb
propriate subcarrier demapping for each of the J simultaneous obtained, for any given N 0
, through an ensemble average
users, allowing subsequent, separate FDE procedures for the (over a large number of simulated channel realizations) of the
several users. Fig. 3.B shows the well-known structure of a conditioned BER, written as
 
conventional FDE scheme for a Q-branch receiver [GDCE00], (q),(in)
P e {Hj ; q = 1, 2, · · · , Q} = (13)
[GDE03], in this case concerning user j. In sec. III, we will v 
Mj −1 Q
consider improved, iterative FDE techniques, by adapting the
u
u 2η Eb X X (q),(in) 2 
approach of [GTDE06], [GTDE07] to this SC-FDMA context. Q t PN −1 Hj,k
Mj n=0 Pn N0 k=0 q=1

C. Uncoded Performance Bounds III. I TERATIVE R ECEIVER T ECHNIQUES


A detailed, insightful evaluation of the iterative receiver A. Simplified Iterative FDE Technique
techniques described in sec. III, in the SC-FDMA context In the following, we consider a low-complexity, iterative
considered in this paper, can benefit from a complementary FDE technique which can be regarded as a simplified ver-
evaluation of the corresponding performance bounds. Some sion of the Turbo FDE technique presented in [GTDE06],

3
(i) (i−1)
[GTDE07], appropriately extended from a single-user context leads to Gj,k = 0 and ρ̂j = 0 in (16): this means
to the SC-FDMA context of this paper. The main simplification that a conventional FDE under the MMSE (Minimum Mean
consists of not using a decoding aid within the iterative FDE Squared Error) criterion is actually carried out in the first iter-
procedures; the additional simplification consists of not trying ation. After several iterations and/or for high SNR, typically
(i−1)
to compensate for possible insufficient-CP conditions. With (i−1) (in) (in)
ρ̂j ≈ 1 and Sj,k ≈ Sj,k , leading to approximately
regards to user j, this FDE scheme, depicted in Fig. 4, should (i)
be compared with the conventional FDE scheme of Fig. 3.B. constant Cj,k coefficients (quasi-matched filtering conditions)
Instead of a fixed vector cj for element-by-element multipli- and a nearly perfect soft cancelation of residual ISI after
(i) that quasi-matched filtering. BER performances very close to
cation, after maximal-ratio combining, we have a vector Cj
the MFB derived in sec. II-C are then achievable. It should
which is upgraded from iteration to iteration (i is the iteration
be noted that we can confirm the MFB performance semi-
number). Additionally, there is a soft cancelation of residual
(i) analytically obtained (as explained in sec. II-C) through a
ISI after that multiplication, through the additive vector Gj ,
Monte Carlo simulation for the receiver technique of Fig. 4,
also upgraded from iteration to iteration. Certainly,
by assuming ideal channel estimation and one iteration only,
Q (i−1) (i)
(i)
X (q),(in) (q),(in)∗
with ρ̂j = 1 in the computation of Cj,k (eq. (16)) and
Cj Yj Ĥj (14) (i) (in)
,(i−1)
(in)
q=1 ideal Gj,k coefficients, i.e., Sj,k = Sj,k in (15).
(i) (in)
= γj Sj + Gaussian N oise + Residual ISI,
(i)
with γj in accordance with (12), when replacing
the fixed Cj,k coefficients by the iteratively adjusted
(i)
Cj,k coefficients. The iteratively adjusted vector
(i)
Gj , for soft cancelation of estimated residual ISI
 i, obviously has components
at iteration

given by
2 
(i) (i) (in) (i) (in) PQ b (q),(in)
Gj,k = E γj Sj,k − Cj,k Sj,k q=1 Hj,k , having
(q),(in) (in) (q),(in) Fig. 4. Simplified iterative FDE technique.
in mind Yj,k = Sj,k Hj,k + Gaussian N oise and
(14). Therefore,
Q
" #
(q),(in) 2 (in) (i−1) B. Turbo FDE Technique

(i) (i) (i)
X
Gj,k = γj − Cj,k Ĥj,k Sj,k , (15)
q=1
By considering a true turbo FDE technique, as depicted
in Fig. 5, we can take advantage of the ”full information”
(i−1) (i−1) (i−1) T on the coded bits which is provided by the SISO (Soft-In
 
(in) (in) (in)
when using Sj = Sj,0 , · · · , Sj,Mj −1 = Soft-Out) decoder. This full information includes an ”extrinsic
 (i−1) (i−1)
T information” contribution (to be added to the LLR’s of the
(in) (in)
FMj sj,0 , · · · sj,Mj −1 to denote the frequency- coded bits at the decoder input), thereby helping the FDE
domain version of the currently available estimate of the process through the resulting upgraded values.
data symbol vector, derived from the preceding iteration.
This confirms the results of [GTDE06], [GTDE07] on the
(i)
additive vector used for soft cancelation of ISI. As to the Cj,k
coefficients, the results of [GTDE06], [GTDE07] can also be
extended to our SC-FDMA context. This means that
(i)
(i) KF
Cj,k =   2  P 2 , (16)
(i−1) Q (q),(in)
α̂j + 1 − ρ̂j q=1 Ĥj,k

N N0 (i)
where αj = Mj σj2
and KF is a normalization factor which
(i−1) Fig. 5. Turbo FDE technique.
(i) (i−1) (in)
leads to γj = 1. As to ρ̂j (and also as to sk ,
(i)
required to compute Gj,k ), we need to use the LLR (Log-
Likelihood Ratio) values concerning the coded bits, as pro- IV. P ERFORMANCE E VALUATION
vided by the preceding iteration [GTDE06], [GTDE07]. Since V. C ONCLUSIONS AND F INAL R EMARKS
there is no ”decoding aid” within the iterative FDE technique
R EFERENCES
of Fig. 4, the required LLR values are simply those previously
obtained, through soft bit demapping, from the equalized [FABSE02] D. Falconer, S. Ariyavisitakul, A. Benyamin-Seeyar, and B. Eid-
(in),(i−1) son. ”Frequency Domain Equalization for Single Carrier Broad-
samples s̃j,n . For i = 1, with no a priori information band Wireless Systems”. IEEE Communications Magazine,
on the coded bits, the LLR values are equal to zero, which CM’02, vol. 40(4):58–66, April 2002.

4
[GDCE00] A. Gusmão, R. Dinis, J. Conceição, and N. Esteves. Comparison
of Two Modulation Choices for Broadband Wireless Commu-
nications. IEEE Vehicular Technology Conference, VTC’00
(Spring), vol. 2:1300–1305, May 2000.
[GDE03] A. Gusmão, R. Dinis, and N. Esteves. On Frequency-domain
Equalization and Diversity Combining for Broadband Wireless
Communications. IEEE Transactions on Communications’03,
vol. 51(7):1029–1033, July 2003.
[GTDE06] A. Gusmão, P. Torres, R. Dinis, and N. Esteves. ”A Class
of Iterative FDE Techniques for Reduced-CP SC-Based Block
Transmission”. IEEE International Symposium on Turbo Codes
and Related Topics, ISTC’06, 71(Session 13), April 2006.
[GTDE07] A. Gusmão, P. Torres, R. Dinis, and N. Esteves. ”A Reduced-
CP Approach to SC/FDE Block Transmission for Broadband
Wireless Communications”. IEEE Transactions on Communi-
cations’07, vol. 55(4):801–809, April 2007.
[SKJ94] H. Sari, G. Karam, and I. Jeanclaude. ”An Analysis of Or-
thogonal Frequency Division Multiplexing for Mobile Radio
Applications”. IEEE Vehicular Technology Conference, VTC’94,
vol. 3:1635–1639, June 1994.
[TG07] P. Torres and A. Gusmão. ”Evaluation of Clipped-OFDM and
SC/FDE Alternatives for Block Transmission Using Iterative
Receiver Techniques”. IEEE International Conference on Com-
munications, ICC’07(Spring), June 2007.

Вам также может понравиться