Beruflich Dokumente
Kultur Dokumente
Indian Journal of Science and Technology, Vol 9(20), DOI: 10.17485/ijst/2016/v9i20/81485, May 2016 ISSN (Online) : 0974-5645
Abstract
In this paper, we proposed an improved recovery algorithm, with efficient pilot placement, for Compressed Sensing
(CS)-based sparse channel estimation based on Sparsity Adaptive Matching Pursuit (SAMP) in Orthogonal Frequency
Division Multiplexing (OFDM) communication systems. The proposed algorithm does not require a priori knowledge of
the sparsity, and to approach the true sparsity, adjusts the step size adaptively. Furthermore, estimation accuracy can be
affected by different measurement matrices in CS, due to different pilot arrangements. It is known that, the Cyclic Difference
Set (CDS) is the optimal setoff pilot locations when the signal is sparse on the unitary Discrete Fourier Transform (DFT)
matrix by minimizing the mutual coherence of the measurement matrix. Based on this, an efficient pilot insertion scheme
is introduced in cases where Cyclic Different Set does not exist. Simulation results in the paper show that the channel
estimation algorithm, with the new pilot placement scheme, which offers a better trade-off between the performances
in terms of Bit Error Rate (BER), Mean Squared Error (MSE) and the complexity, when compared to previous estimation
algorithms.
Keywords: Compressed Sensing, Cyclic Difference Set, Sparse Channel Estimation, Sparsity Adaptive Matching Pursuit
2 Vol 9 (20) | May 2016 | www.indjst.org Indian Journal of Science and Technology
V. Nagendra Babu, V. Adinarayana, K. Muralikrishna and P. Rajesh Kumar
where, ηp is the number of paths, and ηp and Tp are the halting criterion: tolerance є and Γ. Therefore, AS-SAMP
amplitude gain and the delay associated with the pth path, halts when the residual’s norm is smaller than є, in which
respectively. The vector of received signal after Discrete є are set to be the noise energy. Meanwhile, sis decreased
Fourier Transforms (DFT) is expressed as when the energy difference of the reconstructed signal
falls below Γ. Starting with a sufficiently large initial step
Y = XH + W = XDh +W, (7) size (SI ≤ K), the algorithm quickly approaches the target
where, X is an N × N diagonal matrix with the elements signal. However, when the difference in the energy of the
X(k), 0 ≤ k ≤ N − 1, on the main diagonal, Y = [Y(0), reconstructed signals becomes smaller than the pre-set Γ,
Y(1),..., Y(N – 1)]T. H = [H(0), H(1),..., H(N – 1)]T and W the step size is reduced (by a factor of two) to avoid overes-
= [W(0), W(1),..., W(N – 1)]T are the frequency response timation of the K-sparse target signal. This overestimation
vector of the channel and additive white Gaussian noise can significantly degrade the accuracy of the algorithm.
(AWGN), respectively. h = [h(0), h(1),..., h(L – 1)]T. The
1 − j 2pmn/ N , Algorithm-1
(m, n) element of D is given by [D]m,n = e ,
N Algorithm 1 AS-SAMP
where 0 ≤ m ≤ N − 1 and 0 ≤ n ≤ L − 1. After extracting the Input: Received signal at pilot subcarriers Yp, measurement
pilot subcarriers, we can write the following input-output matrix A, tolerance
є, threshold Г, initial step size SI;
relationship 1. Initialize h = [0,0,…,0]T, h old = [0,0,…,0]T, rtemp =
Yp = Xp Dp h + Wp = Ah + Wp, (8) [0,0,…,0]T, indices set D0 = Ø, candidate support set
C0 = Ø,, residual r0 = Yp, size of final support set F0 =
where, Yp = SY, Xp = SXST, Dp = SD, Wp = SW, and S is a P × N Ø,, iteration index t = 1
matrix for selected pilot subcarriers. In addition, A = Xp Dp 2. while (||rt-1|| > є) do
is a P × L matrix, referred to as the measurement matrix. The 3. Calculate signal SP = |AH r(t-1)|
goal of CS-based channel estimation is to estimate h from 4. Select indices set Dt in A corresponding to the L
the received pilot Yp, given the m easurement matrix A. largest elements in SP
{Preliminary test}
3. Proposed AS-SAMP with 5. Merge chosen indices and final support set from
previous iteration into candidate support set Ct = Dt
Application to Sparse Channel Ft–1
Estimation 6. Refine candidate set to final set Ft by selecting indices
corresponding to the L
3.1 Proposed AS-SAMP Algorithm largest elements of | AC t YP | {Final test}
We propose the AS-SAMP algorithm; to expedite the
7. Solve the least-square problem h (Ft) = A F t YP
convergence, the algorithm begins with a larger step size
8. Calculate the current residual rtemp = Yp –
(the initial step size is denoted as SI) which is adaptively
decreased to provide fine tuning in later stages as the A F t A F t YP
change rate of the reconstructed signal’s energy decreases. 9. if (||rtemp|| > є) then
Consequently, an additional threshold Γ is used to specify 10. rt = rt–1
the beginning of the fine tuning. The pseudo code for the 11. Break
proposed algorithm is presented as Algorithm 1. The 12. else if (||rtemp|| ≥ ||rt-1||) then
algorithm is also stage-wise with a variable size of FT in
13. if (|| h (Ft)|| – || h old|| < Г then
different stages. During a stage, it adopts two correlation
tests iteratively, i.e., candidate and final tests, to search a 14. s = [s⁄2], L = L + s, h old = h (Ft), rt = rt–1, t =t + 1 {Fine
certain number of coordinate corresponding to the larg- tuning}
est correlation values between the signal residual and the 15. else
columns of the measurement matrix. Then, the algorithm 16. L = L + s, h old = h (Ft), rt = rt–1, t =t + 1 {Fast
moves to the next stage until the recovered signal with the approaching}
Vol 9 (20) | May 2016 | www.indjst.org Indian Journal of Science and Technology 3
An Improved Compressed Sensing based Sparse Channel Estimation for MIMO-OFDM Systems with an Efficient Pilot Insertion
Scheme
17. end if referred to this as to the iterative tail search; the pseudo
18. else code is shown as Algorithm 2. The size of the search space
19. Ft-1 = Ft, rt = rtemp, t =t + 1 is greatly reduced after concatenation, and hence, the pro-
20. end if posed method converges much faster when compared to
21. end while the iterative methods.
22. return h
Output: Estimation of baseband channel impulse Algorithm – 2
response h
Algorithm 2 Pilot placement Based on Concatenated
CDS with an Iterative tail Search
3.2 Computational Complexity Input: An existing (u, υ, α) CDS C for concatenation, the
In this section, the computational complexity of the existing number of total subcarriers N, the number of pilot sub-
algorithms in the literature and the proposed algorithm is carriers P, the partial DFT matrix D of which the (m, n)
compared in terms of the number of operations. In addi- 1 − j 2pmn/ N
element is e , where 0 ≤ m ≤ N-1, 0 ≤ n ≤
tion, the complexity of the algorithms also depends on the N
number of iterations. A summary of the computational L–1, and L is the number of taps of the CIR;
0
complexity of the considered algorithms OMP, CoSaMP, 1. Initialize Ωc = ∅, Ωtemp =∅
SAMP and AS SAMP is provided in Table 1. N
2. for i from 1 to do
u
4. Pilot Placement i i -1
3. Ωc = {Ωc [C + (I − 1) × u ]}
Suppose that the measurement matrix A is composed of 4. end for N
Prows of the N × L partial DFT matrix D and the indices 5. Pr = P – u × u , Ω = ΩC
set of the selected rows is Ω, and all the pilot symbols are 6. for j from 1 to Pr do
equal-powered. We denote the number of repetitions of
7. Ωtemp = Ω
the different elements of Gas λG= {λg/g = 1, 2..., N − 1},
8. From all Pr –j + 1 possible subsets of size j by adding an
then λ1 = λ2 = ... λ1n–1 = λ which also means that the vari-
element to Ωtemp:
ance of λG is zero. Consider an OFDM system with N=
Ω = {Ωtemp k є {Pr+ 1, Pr+ 2,…, N} \ Ωtemp}
1024, in which P= 256 identical pilot symbols are ran-
9. From the matrix A by selecting rows of D for each j-
domly scattered, and the number of taps of the sparse CIR
element sets generated from the previous step, and the
is L= 250. In order to quantize the channel estimation
indices set of the selected rows is Ω
error, we adopt MSE, which is defined as
10. From all (Pr –j + 1) of A matrices generated from the
∑
N
MSE = E[ | H (m) − H (m)|2 ] (9) previous step, calculate the corresponding mutual
m =1
coherence, and choose the set with the minimum
We propose a pilot placement scheme based on the
mutual coherence
concatenated CDS for pairs of (P, N) where, CDS does
11. end for
not exist. Specifically; we select the existing CDS with the
12. return Ω
parameters (u, υ, α) in which u/υ is the closest to P/N.
Output:The pilot indices set Ω
For instance, to select indices for 256 pilots from 1024
To show that as the variance of λGincreases, it is likely
positions, the (133, 33, 8) CDS is used. After concatenat-
that so does the mutual coherence of A and the MSE of
ing the selected CDS, we adopt an iterative procedure to
estimates, Spearman’s rank correlation is adopted to mea-
find the rest of pilot positions which minimize the mutual
sure the strength of a monotonic relationship (i.e., values
coherence of the resulting measurement matrix. We
of elements in a vector either increase or decrease with
Table 1. Computational complexity every increase in an associated vector) between paired
Methods OMP CoSaMP SAMP AS-SAMP
vectors. Table 2 shows the Spearman’s rank correlation
|h | −1 |h | −1
between any pair of the following four vectors: the vari-
KPN ≤ KPN ≤ [−Jlog( min )
h log(c K s−SAMP )
+ J ] PN ≤ [−Jlog( min )
h log(c K s − ASSAMP )
+ J ] PN
ance of λG, the mutual coherence μ(A), and the MSE
V. Simulation tables for both the OMP and AS-SAMP algorithms, obtained
4 Vol 9 (20) | May 2016 | www.indjst.org Indian Journal of Science and Technology
V. Nagendra Babu, V. Adinarayana, K. Muralikrishna and P. Rajesh Kumar
Vol 9 (20) | May 2016 | www.indjst.org Indian Journal of Science and Technology 5
An Improved Compressed Sensing based Sparse Channel Estimation for MIMO-OFDM Systems with an Efficient Pilot Insertion
Scheme
6. Conclusion
Figure 5. MSE performance of the SAMP and AS SAMP In this paper, we have proposed an adaptive step size
and AS-SAMP algorithms with various number of pilots. SAMP algorithm, AS-SAMP, with an efficient near-opti-
algorithms with different step sizes. mal pilot placement scheme for sparse channel estimation
in OFDM systems. It possesses the advantage of not
times are exponentially distributed with a mean of 1 ms, requiring a priori knowledge of the sparsity of the chan-
i.e., E[τ_(j+1) – τ_j] = 1 ms, j ∈ {0,1,..., N_p – 1}. The nel. It is shown through performance analysis that the
average power decreasing exponentially with delay as the proposed algorithm can significantly improve the estima-
amplitudes are Rayleigh distributed, where the difference tion accuracy without introducing significant additional
between the beginning and the end of the CP is 20 dB. The complexity. In order to ensure a satisfactory estimation,
parameters for the considered reconstruction algorithms we have further proposed a near-optimal pilot placement
are given in Table 3. scheme, which is based on the concatenated CDS with an
We compare the proposed algorithm with two classic iterative tail search. Simulations show that, without sig-
algorithms, namely Least Square (LS) and OMP using dif- nificantly increasing the computational complexity, the
ferent numbers of randomly distributed pilots. Figure 1. proposed AS-SAMP with the new pilot placement scheme
Compares the MSE performance of the LS, OMP and provides a better MSE performance for the channel
AS-SAMP. Figures 2, 3 and 4 plot the MSE, BER and estimate, as well as the system BER.
6 Vol 9 (20) | May 2016 | www.indjst.org Indian Journal of Science and Technology
V. Nagendra Babu, V. Adinarayana, K. Muralikrishna and P. Rajesh Kumar
Vol 9 (20) | May 2016 | www.indjst.org Indian Journal of Science and Technology 7