Академический Документы
Профессиональный Документы
Культура Документы
2
, which is uncorrelated for different n or k.
For the initial Z input OFDM symbols, first L symbols are preambles and the rest are data symbols. The
output Y(n-1,k) for the first preamble X(n-1,k), after removal of CP and taking K-point FFT, is used for first LS
estimate
( 1, ) of the channel as
H
1, =
Y(1,)
X(1,)
(2)
Using similar approach, other LS channel estimates H
( 2, ), H
( 3, ), ..., H
(3)
and a K x 1 reference vector
() is constructed as
= 1,1 1,2 1,
(4)
where [.]
T
represents transpose.
Z-OFDM
Blocks
Update P(n) and
Href(n)
LS Channel Estimation
Initial L
Preambles
X(n,k)
Y(n, k)
2D-RLS/2D-SM-AP
Algorithm
1/ H(n)
X(n,k)
H(n, k) = Y(n, k) / X(n, k)
-1
+1
Decision Quantizer
Fig. 1. Block Diagram Representation of Channel Estimation Method
Using the vectors () and
(), the adaptive channel estimation algorithm is used for estimating the
channel
(). For every next Z OFDM symbols, input is estimated using the previous channel estimate and
Decision Quantizer [6] maps the input estimate to the corresponding BPSK symbols. Using these input
estimated, new L LS estimates of channel are obtained,
().
3. Set-Membership Filtering
The adaptive channel estimation Eq.(2) is given by
( 1)() (5)
where () is LK x K weight coefficient matrix which needs to be estimated by adaptive algorithms and ()
is input vector as described in Eq. (3). The K x 1 priori estimation error vector (n) is given as
=
1() (6)
where
ref
H
}
(,
)
(7)
Given a set of pairs (,
), define (n) as the set containing all matrices such that the associated output
error at time instant n is upper bounded in magnitude by . That is,
= {
ref
H
} (8)
The set () is usually referred as the constraint set [1]. The boundaries of () are hyperplanes. Since for
each pair (,
) there is associated constraint set, the intersection sets over all the available time instants is
called the exact membership set (n), formally defined as
() = ()
=0
(9)
Fig. 2. Exact Membership Set out of Constraint Set Fig. 3.Coefcient matrix updating for the 2D-SM-AP algorithm
The set (n) represents a polygon in the parameter space whose location is one of the main objectives of the
SMF. This polygon in and (n) should become small. In this case, the adaptive-filter coefficients do not need
updating because the current membership set is totally inside the constraint set, resulting in selection of update
which is data dependent. The selective updating of the SMF brings about opportunities for power and
computational savings.
As can be seen from Fig. 2, the feasibility set is a subset of the exact membership set (n). The goal of SMF
adaptive filtering is to adaptively find an estimate that belongs to the feasibility set. 2D-SM-AP algorithm has
been proposed to adaptively obtain the filter coefficients belonging to the feasibility set.
3.1 2D-SM-AP Algorithm
The equation for exact membership set (n) can be expressed as
=
1
=0
=
=
1
()
+1
() (10)
where
+1
() represents the intersection of the (+1) last constraint sets, whereas
1
() is the
intersection of first ( ) constraint sets. The aim of this derivation is to conceive an algorithm whose
coefficient update belongs to the last (+1) constraint sets, i.e.
+1
().
We can retain the last (+1) input vectors in a matrix as follows
= 1 (11)
where is as defined in Eq.(3). In a similar way
() and
2
+1
()]
(12)
where
+1
()()
(13)
() (14)
where is the variable step size and (0 < < 1) is added to avoid large step size when
becomes small.
=
1
()
, () >
0,
(15)
3.2 Algorithm Summary
Step.1: Initialization (n = 0)
Regularization parameter : 0 < < 1
= 5
2
0 =
where
2
represents noise variance.
Step.2: The algorithm operates on a set of X received OFDM symbols at a time. L LS channel estimations are
done using the preamble symbols using equation (2) and P(n),
2
)
6248
2D-SM-AP ( = 0.5
2
)
8591
2D-SM-AP ( = 0.05
2
)
10934
Table 2: Total Operation Count for Convergence for L = 2 and K = 64 at SNR = 10 dB
Table II shows the total operation count required for the convergence of 2D-SM-AP algorithm for different
threshold values and 2D-RLS algorithm at SNR 10dB. It can be seen that as the threshold value is decreased, the
total operation count required for convergence of the proposed algorithm is increased.
Fig. 4 compares the BER performance of 2D-SM-AP algorithm with different threshold values and 2D-RLS
algorithm. Here, CP of length 0.8 s is considered which is greater than T
d
of the channel, hence ISI can be
neglected. It is evident from the figure that the BER performance of both the algorithms is similar. Also,
changing the threshold values does not affect the BER performance of the 2D-SM-AP algorithm.
Fig. 5 shows the Normalized Mean Square Error (NMSE) performance of 2D-SM-AP and 2D-RLS algorithms
at SNR 10dB. It is seen that 2D-RLS converges very fast in only 5 iterations, whereas the convergence speed of
2D-SM-AP algorithms is slow, taking around 10 15 iterations which is acceptable in stationary environments.
However, as the threshold value becomes less, the increased total operation count leads to more number of
iterations required for convergence. Also, the error floor for the algorithms is nearly same.
Fig. 4. BER performance of 2D-RLS & 2D-SM-AP algorithms
Fig. 5. NMSE performance of 2D-RLS and 2D-SM-NLMS algorithms at SNR 10dB
5. Conclusion
In this paper, 2D-SM-N adaptive channel estimation algorithm has been proposed for OFDM systems. The
adaptive-lter coefficients are updated only when the error is more than the predefined threshold. This algorithm
slightly compromises on the convergence speed which is acceptable in stationary environments. But it provides
improvement in the computational complexity of the algorithm (O(N)) than the 2D-RLS adaptive channel
estimation algorithm (O(N
2
)). Reducing the threshold value increases the operation count for convergence
without affecting the BER performance of 2D-SM-AP algorithm. Matlab simulations show that the low
computational complexity is achieved keeping the BER performance of 2D-SM-AP algorithm similar to the
conventional 2D-RLS algorithm.
References
1. P.S.R. Diniz, Adaptive Filtering-Algorithms and Practical Implementation, 3rd ed., Springer, 2008.
2. P.S.R. Diniz, G.O. Pinto and A. Hjrungnes, Data Selective Partial-Update Afne Projection Algorithm,
International Conference on Acoustics, Speech and Signal Processing (ICASSP), March 2008.
3. S. Gollamudi, S. Nagaraj, S. Kapoor, and Y.F. Huang, Set-Membership Filtering and a Set-Membership
Normalized LMS Algorithm with an Adaptive Step Size, IEEE Signal Processing Letters, Vol. 5, pp. 111-
114, May 1998.
4. S. Werner and P.S.R. Diniz,Set-Membership Afne Projection Algorithm, IEEE Signal Processing
Letters, Vol. 8, No. 8, August 2001.
5. T. Sharma, A. Soni and V.K. Chakka,Two-Dimensional Set Membership Normalized Least Mean Square
Adaptive Channel Estimation for OFDM Systems, Proceedings of INDICON 2009, pp. 125-128,
December 2009.
6. X. Hou, S. Li, C. Yin and G. Yue, Two-Dimensional Recursive Least Square Adaptive Channel
Estimation for OFDM Systems, International Conference on Wireless Communications, Networking and
Mobile Computing, pp. 232-236, 2005.
7. P. Hoeher, S. Kaiser and P. Robertson, Two-Dimensional Pilot-Symbol-Aided Channel Estimation by
Wiener Filtering, International Conference on Acoustics, Speech and Signal Processing (ICASSP), 1997.
8. S.K. Kim and Y.H. Lee, 3-Dimentional MMSE Channel Estimation in Multi-Antenna OFDM Systems,
The Third International Conference on Digital Communications, pp. 6-10, 2008.
9. T.S. Rappaport, Wireless Communications, Pearson Edition, 2002.