BEACON Channel Estimation for Cooperative Wireless Sensor Networks Based on Data Selection Tong Wang, Rodrigo C. de Lamare and Paul D. Mitchell Department of Electronics, University of York, UK [email protected] {rcdl500, pdm106}@ohm.york.ac.uk

AbstractβIn this paper, we consider a general cooperative wireless sensor network (WSN) and the problem of channel estimation. A matrix-based set-membership recursive least squares (RLS) algorithm called BEACON is developed for the estimation of the complex channel parameters in order to reduce the computational complexity significantly and extend the lifetime of the WSN by reducing its power consumption. Then, we present and incorporate an error bound function into the BEACON channel estimation method which can adjust the error bound automatically with the update of the channel estimates. Computer simulations show good performance of our proposed algorithms in terms of convergence speed and steady state mean square error, reduced complexity and robustness to the time-varying environment and different signal-to-noise ratio (SNR) values.

I. I NTRODUCTION Recently, there has been a growing research interest in wireless sensor networks (WSNs) because their unique features allow a wide range of applications [1]. They are usually composed of a large number of densely deployed sensing devices which can transmit their data to the desired user through multihop relays [2]. Low complexity and high energyefficiency are the most important design characteristics of communication protocols [3] and physical layer techniques. The performance and capacity of WSNs can be significantly enhanced through exploitation of spatial diversity with cooperation between the nodes [2]. In a cooperative WSN, nodes relay signals to each other in order to propagate redundant copies of the same signals to the destination nodes. Among the existing relay schemes, amplify-and-forward (AF) and the decode-and-forward (DF) are the most popular approaches [4]. Due to limitations in sensor node power, computational capacity and memory [1], some power-constrained relay strategies [5], [6] and power allocation methods [7] have been proposed for WSNs to obtain the best possible SNR or best possible quality of service (QoS) at the destinations. Most of these ideas are based on the assumption of perfect synchronization and available channel state information (CSI) at each node [1]. Therefore, more accurate estimates of the CSI will bring about better performance in WSNs. The least mean squares (LMS) and normalized least mean squares (NLMS) estimation methods are appropriate for WSNs due to their simplicity. Moreover, a set-membership NLMS (SM-NLMS) channel estimator for WSNs has been proposed in [8] which outperforms the conventional NLMS channel estimation offering reduced computational complexity. Compared with the LMS and NLMS channel estimation methods, the RLS channel estimator can provide better performance in terms of the convergence speed and steady state

978-1-4244-6317-6/10/$26.00 Β© 2010 IEEE

[9]. However, it is not suitable for the WSNs due to its high computational complexity [9]. In order to overcome this shortcoming, the set-membership filtering (SMF) framework [10] can be introduced to propose a computationally efficient version of the conventional RLS channel estimation method, called BEACON channel estimation. It can be considered as a constrained optimization problem where the objective function is the least squares (LS) cost function and the constraint is a bound on the magnitude of the estimation error. As a result, an adaptive forgetting factor can be derived to achieve the optimal performance for each update. Most importantly, the algorithm possesses a feature that allows updating for only a small fraction of the time, expressed as the update rate (UR). The UR of the BEACON channel estimation decreases obviously due to the data-selective update which can reduce the computational complexity significantly and extend the lifetime of the WSN by reducing its power consumption. The biggest issue for BEACON channel estimation is appropriate selection of the error bound, because it has a critical effect on the estimation performance. The value of error bound can be varied to trade off between achievable performance and computational complexity [11]. A higher error bound would result in lower UR but worse performance. For WSNs the aim is to achieve an acceptable CSI quickly with low power consumption. Therefore, the bound should be adjusted to make the estimation performance prior to the computational complexity during the first updates and then lower the UR gradually. Also, the required error bound may be time variant due to changing environmental conditions. In this paper, we develop a matrix-based BEACON algorithm for channel estimation in cooperative WSNs using the AF cooperation protocol. The major novelty in the BEACON algorithms presented here is that they are matrix-based channel estimation algorithms as opposed to vector-based BEACON techniques for filtering applications [11], [12]. Therefore we specify a bound on the norm of the estimation error vector instead of the magnitude of the scalar estimation error. Then, a novel error bound function is introduced to change the error bound automatically in order to obtain optimal performance with the proposed BEACON channel estimation. A key contribution of this paper is the consideration of techniques to reduce the complexity of the channel estimation for WSNs. This paper is organized as follows. Section II describes the general cooperative WSN system model and its constrained form. Section III introduces the BEACON channel estimation method using the SMF framework and presents an error bound function which tunes the error bound automatically. Section IV presents and discusses the simulation results, while Section V provides some concluding remarks.

140

ISWCS 2010

Hd

II. C OOPERATIVE WSN S YSTEM M ODEL Consider a general m-hop wireless sensor network (WSN) with multiple parallel relay nodes for each hop, as shown in Fig. 1. The WSN consists of ππ sources, ππ destinations and ππ relays which are separated into πβ1 groups: ππ(1) ,ππ(2) , ... ,ππ(πβ1) . All these nodes are assumed to be within the communication range. We will concentrate on a time division scheme with perfect synchronization, for which all signals are transmitted and received in separate time slots. The sources first broadcast the ππ Γ 1 signal vector s to the destinations and all groups of relays. We consider an amplify-and-forward (AF) cooperation protocol in this paper. Each group of relays receives the signal from the sources and previous groups of relays, amplifies and rebroadcasts them to the next groups of relays and the destinations. In practice, we need to consider the constraints on the transmission policy. For example, each emitter would transmit during only one phase. In our WSN system, we assume that each group of relays transmits the signal to the nearest group of relays and the destinations directly. We can use a block diagram to indicate the cooperative WSN system with these transmission constraints as shown in Fig. 2.

Sources Ns

Cooperative Relays Nr

Destinations

d

Hr(m-1),d

dm

xm-1

vr(m-1) H

vdm-1 dm-1

Hr(m-2),d

r(m-2),r(m-1)

xm-2 x2

vr(2)

v d2

Hr(1),r(2)

d2

Hr(1),d

x1

vr(1)

vd

1

d1

Hs,d

Hs,r(1) s

Fig. 2. Block diagram of the cooperative WSN system with transmisssion constraints.

.. .

.. .

Nd

v dm

Phase π: (π = 2, 3, ..., π β 1) xπ = Hπ(πβ1),π(π) Aπβ1 xπβ1 + vπ(π)

(5)

dπ = Hπ(πβ1),π Aπβ1 xπβ1 + vππ

(6)

dπ = Hπ(πβ1),π Aπβ1 xπβ1 + vπ π

(7)

Phase π:

where v is a zero-mean circularly symmetric complex additive white Gaussian noise (AWGN) vector with covariance matrix π 2 I. Aπ is a diagonal matrix whose elements represent the amplification coefficient of each relay of the πth group. The vectors dπ and vππ denote the received signal and noise at the destinations during the πth phase, respectively. At the destinations, the received signal can be expressed as: d = Hπ Ay + vπ Fig. 1. An π-hop cooperative WSN with ππ sources, ππ destinations and ππ relays.

where β‘

Let Hπ ,π(π) denotes the ππ(π) Γ ππ channel matrix between the sources and the πth group of relays, Hπ(π),π denotes the ππ Γ ππ(π) channel matrix between the πth group of relays and destinations, and Hπ(πβ1),π(π) denotes the ππ(π) Γ ππ(πβ1) channel matrix between two groups of relays. We consider a quasi-static fading channel and assume that all the channels in this system fade independently. The received signal at the πth group of relays (xπ ) and destinations (d) for each phase can be expressed as: Phase 1: x1 = Hπ ,π(1) s + vπ(1)

(1)

d1 = Hπ ,π s + v1π

(2)

x2 = Hπ(1),π(2) A1 x1 + vπ(2)

(3)

d2 = Hπ(1),π A1 x1 + v2π

(4)

Phase 2:

141

(8) β‘

β‘

β€ dπ β’β β β β₯ β’ πβ1 β₯ β’d β₯ β’β β β β₯ β’ β₯ β’ . β₯ β₯ . d=β’ β’ . β₯, β’β β β β₯ β’ β₯ β’ d2 β₯ β’ β₯ β£β β β β¦

β€ vπ π β’β β ββ₯ β’ πβ1 β₯ β’ vπ β₯ β’ β₯ β’β β ββ₯ β’ . β₯ β₯ vπ = β’ β’ .. β₯ , β’ β₯ β’β β ββ₯ β’ 2 β₯ β’ vπ β₯ β£β β ββ¦ v1π

β€ xπβ1 β’β β ββ₯ β’ β₯ β’ xπβ2 β₯ β’ β₯ β’β β ββ₯ β’ . β₯ β₯ y=β’ β’ .. β₯ , (9) β’ β₯ β’β β ββ₯ β’ x β₯ β’ 1 β₯ β£β β ββ¦ s

(πππ Γ 1)

(πππ Γ 1)

((ππ + ππ ) Γ 1)

d1

β‘ β’ β’ Hπ = β’ β’ β£

Hπ(πβ1),π .. . 0

β β β

0

..

.. .

Hπ(πβ2),π .

β β β (πππ Γ (ππ + ππ ))

Hπ(1),π Hπ ,π

β€ β₯ β₯ β₯ β₯ β¦ (10)

β‘ β’ β’ A=β’ β’ β£

β β β

Aπβ1

0

β₯ .. β₯ β₯ . β₯ β¦

Aπβ2

.. .

..

0

. A1

β β β

To solve this constrained optimization problem, we can modify the LS cost function using the method of Lagrange multipliers which yields the following Lagrangian function:

β€ (11)

π½=

I

πβ1 β

π(π)πβπ β£β£r(π) β H(π)s(π)β£β£2

π=1

((ππ + ππ ) Γ (ππ + ππ )) Here, we use dashed lines to separate the vectors d, vπ and y in order to distinguish between transmissions to the destinations in π different time slots. The matrix Hπ consists of all the channels between each group of relays and destinations. The matrix A consists of the amplification coefficients of all relays.

[

2

+ π(π) β£β£r(π) β H(π)s(π)β£β£ β πΎ

2

]

where π(π) plays the role of both the Lagrange multiplier and the forgetting factor of the LS cost function. By setting the gradient of π½ with respect to H(π) equal to zero, we get: ] [πβ1 β πβπ π» π» π(π) r(π)s (π) + π(π)r(π)s (π) H(π) = π=1

III. S ET-M EMBERSHIP C HANNEL E STIMATION Consider a channel estimation problem where the estimation error is defined as: e = r β Hs, (12) where s is the pilot signal vector, H is the estimated channel matrix and r is the received signal vector at the destination. Conventional channel estimation schemes seek to find the channel matrix H by minimizing a cost function which is a suitable objective function of the estimation error vector e. For example, a least squares channel estimation minimizes the weighted sum of the squared norm of the error vector β£β£eβ£β£2 . In contrast, set-membership (SM) channel estimation specifies an upper bound πΎ on the norm of the estimation error vector over a model space of interest which is denoted as π, comprising all possible pilot-received signal pairs (s, r). The SM criterion corresponds to finding H that satisfies: 2

2

β£β£e(H)β£β£ β€ πΎ , β(s, r) β π.

(13)

The set of all possible H that satisfy (13) is referred to as the feasibility set and can be expressed as: β© { } H β πΆ π : β£β£r β Hsβ£β£ β€ πΎ . (14) Ξ=

(17)

Γ

[πβ1 β

]β1 π(π)

πβπ

π»

π»

s(π)s (π) + π(π)s(π)s (π)

π=1

(18)

Let π(π) =

πβ1 β

π(π)πβπ s(π)sπ» (π) + π(π)s(π)sπ» (π)

(19)

π(π)πβπ r(π)sπ» (π) + π(π)r(π)sπ» (π)

(20)

π=1

and Z(π) =

πβ1 β π=1

Eqation (18) becomes: H(π) = Z(π)πβ1 (π)

(21)

Isolating the term corresponding to π = π β 1 from the rest of the summation on the right-hand side of (19), we may write: ] [πβ2 β πβπ π» π» π(π) s(π)s (π) + π(π)s(π β 1)s (π β 1) π(π) = π=1

+ π(π)s(π)sπ» (π) (22)

(s,r)βπ

At time instant π, the constraint set πΆπ is defined as the set of all H(π) that satisfy (13) for the pilot-received signal pairs (s(π), r(π)): { } (15) πΆπ = H(π) β πΆ π : β£β£r(π) β H(π)s(π)β£β£ β€ πΎ .

The expression inside the brackets on the right-hand side of (22) equals π(πβ1) assuming the forgetting factor of the cost function is close to 1. Hence, we have the following recursion for updating the value of π(π):

The idea behind the SM channel estimation is that if the estimated channel at time instant π β 1 (H(π β 1)) lies outside the constraint set πΆπ , i.e. β£β£r(π) β H(π β 1)s(π)β£β£ > πΎ, we will make the estimated channel at the next time instant (H(π)) lie on the closest boundary of πΆπ . Otherwise, there is no need to compute and the consumption of power can be greatly saved.

Similarly, we may use (20) to derive the following recursion for updating Z(π):

π(π) = π(π β 1) + π(π)s(π)sπ» (π)

Z(π) = Z(π β 1) + π(π)r(π)sπ» (π)

The proposed BEACON channel estimation method can be considered as the following optimization problem: minimize

πβ1 β π=1

π(π)πβ1 (π β 1)s(π)sπ» (π)π(π)πβ1 (π β 1) 1 + π(π)sπ» (π)πβ1 (π β 1)s(π) (25)

For convenience of computation, let: P(π) = πβ1 (π)

π(π)

πβπ

β£β£r(π) β H(π)s(π)β£β£ 2

subject to β£β£r(π) β H(π)s(π)β£β£ = πΎ

2

(16)

(24)

Then, using the matrix inversion lemma [9], we obtain the following recursive equation for the inverse of π(π): πβ1 (π) = πβ1 (πβ1)β

A. Proposed BEACON Channel Estimation

(23)

and

2

142

k(π) =

sπ» (π)P(π β 1) 1 + π(π)sπ» (π)P(π β 1)s(π)

(26) (27)

TABLE I S UMMARY OF THE BEACON C HANNEL E STIMATION A LGORITHM

Therefore, we may rewrite (21) and (25) as: H(π) = Z(π)P(π)

(28)

P(π) = P(π β 1) β π(π)P(π β 1)s(π)k(π)

(29)

Initialize the algorithm by setting H(0) = 0 P(0) = I For each instant of time, π=1, 2, ..., compute { π(π) = ( r(π) β H(π)β 1)s(π) β£β£π(π)β£β£ 1 β 1 , if β£β£π(π)β£β£ > πΎ, πΊ(π) πΎ π(π) = 0, otherwise. where πΊ(π) = sπ» (π)P(π β 1)s(π)

Then we substitute (24) and (29) into (28) to obtain a recursive equation for updating the channel matrix H(π): H(π) = H(π β 1) β π(π)H(π β 1)s(π)k(π) + π(π)r(π)sπ» (π)P(π)

(30)

sπ» (π)P(πβ1)

k(π) = 1+π(π)πΊ(π) H(π) = H(π β 1) + π(π)π(π)k(π) P(π) = P(π β 1) β π(π)P(π β 1)s(π)k(π)

By rearranging (27) , we can get: k(π) = sπ» (π)P(π β 1) β π(π)sπ» (π)P(π β 1)s(π)k(π) = sπ» (π) [P(π β 1) β π(π)P(π β 1)s(π)k(π)]

B. BEACON Channel Estimation with a Time-Varying Bound

= sπ» (π)P(π) (31) Using (31) above, we get the desired recursive equation for updating the channel matrix H(π): H(π) = H(π β 1) β π(π)H(π β 1)s(π)k(π) + π(π)r(π)k(π) = H(π β 1) + π(π) [r(π) β H(π β 1)s(π)] k(π) = H(π β 1) + π(π)π(π)k(π) (32) where π(π) = r(π)βH(πβ1)s(π) denotes the prediction error vector at time instant π. Then we can get the error vector: e(π) = r(π) β H(π)s(π)

(33)

By substituting (32) into (33), we have: e(π) = r(π) β [H(π β 1) + π(π)π(π)k(π)] s(π) = r(π) β H(π β 1)s(π) β π(π)π(π)k(π)s(π) sπ» (π)P(π β 1)s(π) = π(π) β π(π)π(π) 1 + π(π)sπ» (π)P(π β 1)s(π) πΊ(π) (34) = π(π) β π(π)π(π) 1 + π(π)πΊ(π) [ ] π(π)πΊ(π) = π(π) 1 β 1 + π(π)πΊ(π) 1 = π(π) 1 + π(π)πΊ(π) where πΊ(π) = sH (π)P(π β 1)s(π). The constraint set is described as: 1 β£β£ β€ πΎ (35) β£β£e(π)β£β£ = β£β£π(π) 1 + π(π)πΊ(π) If β£β£π(π)β£β£ > πΎ, then the previous solution lies outside the constraint set. We can choose the constraint value β£β£e(π)β£β£ equal to πΎ so that the new solution lies on the closest boundary of the constraint set. Therefore: 1 = πΎ. (36) β£β£e(π)β£β£ = β£β£π(π)β£β£ β£1 + π(π)πΊ(π)β£ Hence the optimal forgetting factor at the πth iteration can be expressed as: ) ( β£β£π(π)β£β£ 1 β1 (37) π(π) = πΊ(π) πΎ Table I shows a summary of the BEACON channel estimation algorithm which will be used for the simulations.

In order to obtain the optimal error bound at each time instant, in this section we introduce an error bound function which can adjust the error bound automatically with the update of the channel estimate. It can be expressed as: β (38) πΎ(π + 1) = (1 β π½)πΎ(π) + π½ πΌβ£β£H(π)β£β£2 π 2 , where π½ is the forgetting factor, πΌ is the tuning parameter, and π 2 is the variance of the noise which is assumed to be known at the destinations. IV. S IMULATIONS In this section, we numerically study the performance of our proposed BEACON channel estimation method as well as the design of the optimal error bound. We consider a 3-hop wireless sensor network. The number of sources (ππ ), two groups of relays (ππ(1) , ππ(2) ) and destinations (ππ ) are 2, 4, 4, 3 respectively. We consider an AF cooperation protocol and the amplification coefficient of each relay is set to 1 for the purpose of simplification. We choose Hπ as our estimated channel because it is the most significant and most complex channel among all channels of the WSN system. During each phase, the sources and each group of relays transmit the QPSK modulated samples in the form of packets containing 2000 pilot symbols. The noise at the destinations is modeled as circularly symmetric complex Gaussian random variables with zero mean. The SNR is fixed at 10 dB. Fig. 3 shows the mean square error (MSE) performance of our proposed BEACON channel estimation, and compares it with conventional RLS channel estimation. Also, the minimum-mean-square error (MMSE) channel estimator which requires the full a priori knowledge of the channel correlation matrix and the noise variance is used here for reference. For the BEACON estimator, we choose four fixed error bounds (πΎ) ranging from 0.6 to 0.9. It can be seen that a higher value of πΎ results in worse MSE performance but a lower update rate (UR). It means the update is selective which can reduce the computational complexity and power consumption. In the case of an error bound equal to 0.6, the BEACON algorithm outperforms the conventional RLS algorithm (with a forgetting factor of 0.998) in terms of convergence speed and steady state with a slightly reduced UR (0.9128). When the error bound is increased to 0.8, although its convergence speed is slower than RLS channel estimation, the final MSE is comparable with a much lower UR (0.4356).

143

0

β2

10

10

RLS Ξ»=0.998 BEACON Ξ³=0.6 UR=0.9128 BEACON Ξ³=0.7 UR=0.7133 BEACON Ξ³=0.8 UR=0.4356 BEACON Ξ³=0.9 UR=0.2228 MMSE

β1

β3

10

β2

Mean Square Error

Mean Square Error

10

RLS Ξ»=0.998 BEACON Ξ³=0.6 BEACON Ξ³=0.7 BEACON Ξ³=0.8 BEACON Ξ³=0.9 BEACON with TVB Ξ±=3 Ξ²=0.001 UR=0.4003

10

β3

10

β4

10

β5

β4

10

10

β5

10

β6

10

0

500

1000 Received Symbols

1500

2000

0

2

4

6

8

10 SNR (dB)

12

14

16

18

20

Fig. 3. MSE performance of the BEACON channel estimation of Hπ compared with the RLS channel estimation.

Fig. 5. Channel estimation MSEs versus SNR for both the fixed bound and time-varying bound.

Fig. 4 illustrates the performance when we apply the timevarying bound (TVB) into the BEACON channel estimation. We set πΌ to 3 and π½ to 0.001. Our proposed algorithm can achieve very similar performance to the conventional RLS channel estimation with a substantial reduction in the UR. Therefore, the computational complexity is significantly reduced which is our goal.

V. C ONCLUSIONS We have proposed a BEACON channel estimation method based on a time-varying bound for cooperative wireless sensor networks. It has been shown that our proposed method can achieve similar performance to conventional RLS channel estimation, offering reduced computational complexity. Furthermore, the incorporation of the time-varying bound function makes it robust to changes in the environment. These features are desirable for WSNs and bring about a significant reduction in energy consumption.

0

10

RLS Ξ»=0.998 BEACON with TVB Ξ±=3 Ξ²=0.001 UR=0.4003 MMSE

β1

Mean Square Error

10

R EFERENCES

β2

10

β3

10

β4

10

β5

10

0

500

1000 Received Symbols

1500

2000

Fig. 4. MSE performance of the BEACON channel estimation with a timevarying bound.

In order to test our proposed channel estimation algorithm in a time-varying environment, the MSE versus SNR performance of the BEACON channel estimation methods is displayed with fixed error bounds and the proposed time-varying error bound in Fig. 5. In the cases of fixed error bounds, the MSE is lower bounded at different values for different error bounds. When the SNR is larger than a specified value, their MSEs will become worse. However, when the time-varying error bound is applied to the BEACON channel estimation, it can be observed that the MSE keeps on decreasing alone with the increase of the SNR. It shows the robustness to the time-varying environment of our proposed algorithm.

144

[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, βA Survey on Sensor Networks,β IEEE Communications Magazine, vol. 40, pp. 102-114, August 2002. [2] J. N. Laneman and G. W. Wornell, βCooperative diversity in wireless networks: Efficient protocols and outage behavior,β IEEE Transactions on Information Theory, vol. 50, no. 12, pp. 3062-3080, December 2004. [3] H. Li and P. D. Mitchell, βReservation packet medium access control for wireless sensor networks,β IEEE Personal, Indoor and Mobile Radio Communications Conference (PIMRC), 2008. [4] Y. W. Hong, W. J. Huang, F. H. chiu, and C. C. J. Kuo, βCooperative Communications in Resource-Constrained Wireless Networks,β IEEE Signal Processing Magazine, vol. 24, pp. 47-57, May 2007. [5] N. Khajehnouri and A.H. Sayed, βDistributed MMSE Relay Strategies for Wireless Sensor Networks,β IEEE Transactions on Signal Processing, vol. 55, no.7, pp. 3336-3348, July 2007. [6] R. Krishna, Z. Xiong, and S. Lambotharan, βA Cooperative MMSE Relay Strategy for Wireless Sensor Networks,β IEEE Signal Processing Letters, vol. 15, pp. 549-552, 2008. [7] Y. Li, B. Vucetic, Z. Zhou, and M. Dohler, βDistributed Adaptive Power Allocation for Wireless Relay Networks,β IEEE Transactions on Wireless Communications, vol. 6, no. 3, pp. 948-958, March 2007. [8] T. Wang, R. C. de Lamare and P. D. Mitchell, βLow-Complexity Channel Estimation for Cooperative Wireless Sensor Networks Based on Data Selection,β IEEE Vehicular Technology Conference (VTC), May 2010. [9] S. Haykin, Adaptive Filter Theory, 4th ed. Englewood Cliffs, NJ: Prentice-Hall, 2002. [10] R. C. de Lamare and P. S. R. Diniz, βSet-Membership Adaptive Algorithms Based on Time-Varying Error Bounds for CDMA Interference Suppression,β IEEE Transactions on Vehicular Technology, vol. 58, no. 2, pp. 644-654, February 2009. [11] S. Nagaraj, S. Gollamudi, S. Kapoor, and Y. F. Huang, βBEACON: An Adaptive Set-Membership Filtering Technique with Sparse Updatesβ IEEE Transactions on Signal Processing, vol. 47, pp. 2928-2941, Nov. 1999. [12] S. Dasgupta and Y. F. Huang, βAsymptotically Convergent Modified Recursive Least-Squares with Data Dependent Updating and Forgetting Factorβ 24th IEEE Conference on Decision and Control, pp. 10671071, 1985.

AbstractβIn this paper, we consider a general cooperative wireless sensor network (WSN) and the problem of channel estimation. A matrix-based set-membership recursive least squares (RLS) algorithm called BEACON is developed for the estimation of the complex channel parameters in order to reduce the computational complexity significantly and extend the lifetime of the WSN by reducing its power consumption. Then, we present and incorporate an error bound function into the BEACON channel estimation method which can adjust the error bound automatically with the update of the channel estimates. Computer simulations show good performance of our proposed algorithms in terms of convergence speed and steady state mean square error, reduced complexity and robustness to the time-varying environment and different signal-to-noise ratio (SNR) values.

I. I NTRODUCTION Recently, there has been a growing research interest in wireless sensor networks (WSNs) because their unique features allow a wide range of applications [1]. They are usually composed of a large number of densely deployed sensing devices which can transmit their data to the desired user through multihop relays [2]. Low complexity and high energyefficiency are the most important design characteristics of communication protocols [3] and physical layer techniques. The performance and capacity of WSNs can be significantly enhanced through exploitation of spatial diversity with cooperation between the nodes [2]. In a cooperative WSN, nodes relay signals to each other in order to propagate redundant copies of the same signals to the destination nodes. Among the existing relay schemes, amplify-and-forward (AF) and the decode-and-forward (DF) are the most popular approaches [4]. Due to limitations in sensor node power, computational capacity and memory [1], some power-constrained relay strategies [5], [6] and power allocation methods [7] have been proposed for WSNs to obtain the best possible SNR or best possible quality of service (QoS) at the destinations. Most of these ideas are based on the assumption of perfect synchronization and available channel state information (CSI) at each node [1]. Therefore, more accurate estimates of the CSI will bring about better performance in WSNs. The least mean squares (LMS) and normalized least mean squares (NLMS) estimation methods are appropriate for WSNs due to their simplicity. Moreover, a set-membership NLMS (SM-NLMS) channel estimator for WSNs has been proposed in [8] which outperforms the conventional NLMS channel estimation offering reduced computational complexity. Compared with the LMS and NLMS channel estimation methods, the RLS channel estimator can provide better performance in terms of the convergence speed and steady state

978-1-4244-6317-6/10/$26.00 Β© 2010 IEEE

[9]. However, it is not suitable for the WSNs due to its high computational complexity [9]. In order to overcome this shortcoming, the set-membership filtering (SMF) framework [10] can be introduced to propose a computationally efficient version of the conventional RLS channel estimation method, called BEACON channel estimation. It can be considered as a constrained optimization problem where the objective function is the least squares (LS) cost function and the constraint is a bound on the magnitude of the estimation error. As a result, an adaptive forgetting factor can be derived to achieve the optimal performance for each update. Most importantly, the algorithm possesses a feature that allows updating for only a small fraction of the time, expressed as the update rate (UR). The UR of the BEACON channel estimation decreases obviously due to the data-selective update which can reduce the computational complexity significantly and extend the lifetime of the WSN by reducing its power consumption. The biggest issue for BEACON channel estimation is appropriate selection of the error bound, because it has a critical effect on the estimation performance. The value of error bound can be varied to trade off between achievable performance and computational complexity [11]. A higher error bound would result in lower UR but worse performance. For WSNs the aim is to achieve an acceptable CSI quickly with low power consumption. Therefore, the bound should be adjusted to make the estimation performance prior to the computational complexity during the first updates and then lower the UR gradually. Also, the required error bound may be time variant due to changing environmental conditions. In this paper, we develop a matrix-based BEACON algorithm for channel estimation in cooperative WSNs using the AF cooperation protocol. The major novelty in the BEACON algorithms presented here is that they are matrix-based channel estimation algorithms as opposed to vector-based BEACON techniques for filtering applications [11], [12]. Therefore we specify a bound on the norm of the estimation error vector instead of the magnitude of the scalar estimation error. Then, a novel error bound function is introduced to change the error bound automatically in order to obtain optimal performance with the proposed BEACON channel estimation. A key contribution of this paper is the consideration of techniques to reduce the complexity of the channel estimation for WSNs. This paper is organized as follows. Section II describes the general cooperative WSN system model and its constrained form. Section III introduces the BEACON channel estimation method using the SMF framework and presents an error bound function which tunes the error bound automatically. Section IV presents and discusses the simulation results, while Section V provides some concluding remarks.

140

ISWCS 2010

Hd

II. C OOPERATIVE WSN S YSTEM M ODEL Consider a general m-hop wireless sensor network (WSN) with multiple parallel relay nodes for each hop, as shown in Fig. 1. The WSN consists of ππ sources, ππ destinations and ππ relays which are separated into πβ1 groups: ππ(1) ,ππ(2) , ... ,ππ(πβ1) . All these nodes are assumed to be within the communication range. We will concentrate on a time division scheme with perfect synchronization, for which all signals are transmitted and received in separate time slots. The sources first broadcast the ππ Γ 1 signal vector s to the destinations and all groups of relays. We consider an amplify-and-forward (AF) cooperation protocol in this paper. Each group of relays receives the signal from the sources and previous groups of relays, amplifies and rebroadcasts them to the next groups of relays and the destinations. In practice, we need to consider the constraints on the transmission policy. For example, each emitter would transmit during only one phase. In our WSN system, we assume that each group of relays transmits the signal to the nearest group of relays and the destinations directly. We can use a block diagram to indicate the cooperative WSN system with these transmission constraints as shown in Fig. 2.

Sources Ns

Cooperative Relays Nr

Destinations

d

Hr(m-1),d

dm

xm-1

vr(m-1) H

vdm-1 dm-1

Hr(m-2),d

r(m-2),r(m-1)

xm-2 x2

vr(2)

v d2

Hr(1),r(2)

d2

Hr(1),d

x1

vr(1)

vd

1

d1

Hs,d

Hs,r(1) s

Fig. 2. Block diagram of the cooperative WSN system with transmisssion constraints.

.. .

.. .

Nd

v dm

Phase π: (π = 2, 3, ..., π β 1) xπ = Hπ(πβ1),π(π) Aπβ1 xπβ1 + vπ(π)

(5)

dπ = Hπ(πβ1),π Aπβ1 xπβ1 + vππ

(6)

dπ = Hπ(πβ1),π Aπβ1 xπβ1 + vπ π

(7)

Phase π:

where v is a zero-mean circularly symmetric complex additive white Gaussian noise (AWGN) vector with covariance matrix π 2 I. Aπ is a diagonal matrix whose elements represent the amplification coefficient of each relay of the πth group. The vectors dπ and vππ denote the received signal and noise at the destinations during the πth phase, respectively. At the destinations, the received signal can be expressed as: d = Hπ Ay + vπ Fig. 1. An π-hop cooperative WSN with ππ sources, ππ destinations and ππ relays.

where β‘

Let Hπ ,π(π) denotes the ππ(π) Γ ππ channel matrix between the sources and the πth group of relays, Hπ(π),π denotes the ππ Γ ππ(π) channel matrix between the πth group of relays and destinations, and Hπ(πβ1),π(π) denotes the ππ(π) Γ ππ(πβ1) channel matrix between two groups of relays. We consider a quasi-static fading channel and assume that all the channels in this system fade independently. The received signal at the πth group of relays (xπ ) and destinations (d) for each phase can be expressed as: Phase 1: x1 = Hπ ,π(1) s + vπ(1)

(1)

d1 = Hπ ,π s + v1π

(2)

x2 = Hπ(1),π(2) A1 x1 + vπ(2)

(3)

d2 = Hπ(1),π A1 x1 + v2π

(4)

Phase 2:

141

(8) β‘

β‘

β€ dπ β’β β β β₯ β’ πβ1 β₯ β’d β₯ β’β β β β₯ β’ β₯ β’ . β₯ β₯ . d=β’ β’ . β₯, β’β β β β₯ β’ β₯ β’ d2 β₯ β’ β₯ β£β β β β¦

β€ vπ π β’β β ββ₯ β’ πβ1 β₯ β’ vπ β₯ β’ β₯ β’β β ββ₯ β’ . β₯ β₯ vπ = β’ β’ .. β₯ , β’ β₯ β’β β ββ₯ β’ 2 β₯ β’ vπ β₯ β£β β ββ¦ v1π

β€ xπβ1 β’β β ββ₯ β’ β₯ β’ xπβ2 β₯ β’ β₯ β’β β ββ₯ β’ . β₯ β₯ y=β’ β’ .. β₯ , (9) β’ β₯ β’β β ββ₯ β’ x β₯ β’ 1 β₯ β£β β ββ¦ s

(πππ Γ 1)

(πππ Γ 1)

((ππ + ππ ) Γ 1)

d1

β‘ β’ β’ Hπ = β’ β’ β£

Hπ(πβ1),π .. . 0

β β β

0

..

.. .

Hπ(πβ2),π .

β β β (πππ Γ (ππ + ππ ))

Hπ(1),π Hπ ,π

β€ β₯ β₯ β₯ β₯ β¦ (10)

β‘ β’ β’ A=β’ β’ β£

β β β

Aπβ1

0

β₯ .. β₯ β₯ . β₯ β¦

Aπβ2

.. .

..

0

. A1

β β β

To solve this constrained optimization problem, we can modify the LS cost function using the method of Lagrange multipliers which yields the following Lagrangian function:

β€ (11)

π½=

I

πβ1 β

π(π)πβπ β£β£r(π) β H(π)s(π)β£β£2

π=1

((ππ + ππ ) Γ (ππ + ππ )) Here, we use dashed lines to separate the vectors d, vπ and y in order to distinguish between transmissions to the destinations in π different time slots. The matrix Hπ consists of all the channels between each group of relays and destinations. The matrix A consists of the amplification coefficients of all relays.

[

2

+ π(π) β£β£r(π) β H(π)s(π)β£β£ β πΎ

2

]

where π(π) plays the role of both the Lagrange multiplier and the forgetting factor of the LS cost function. By setting the gradient of π½ with respect to H(π) equal to zero, we get: ] [πβ1 β πβπ π» π» π(π) r(π)s (π) + π(π)r(π)s (π) H(π) = π=1

III. S ET-M EMBERSHIP C HANNEL E STIMATION Consider a channel estimation problem where the estimation error is defined as: e = r β Hs, (12) where s is the pilot signal vector, H is the estimated channel matrix and r is the received signal vector at the destination. Conventional channel estimation schemes seek to find the channel matrix H by minimizing a cost function which is a suitable objective function of the estimation error vector e. For example, a least squares channel estimation minimizes the weighted sum of the squared norm of the error vector β£β£eβ£β£2 . In contrast, set-membership (SM) channel estimation specifies an upper bound πΎ on the norm of the estimation error vector over a model space of interest which is denoted as π, comprising all possible pilot-received signal pairs (s, r). The SM criterion corresponds to finding H that satisfies: 2

2

β£β£e(H)β£β£ β€ πΎ , β(s, r) β π.

(13)

The set of all possible H that satisfy (13) is referred to as the feasibility set and can be expressed as: β© { } H β πΆ π : β£β£r β Hsβ£β£ β€ πΎ . (14) Ξ=

(17)

Γ

[πβ1 β

]β1 π(π)

πβπ

π»

π»

s(π)s (π) + π(π)s(π)s (π)

π=1

(18)

Let π(π) =

πβ1 β

π(π)πβπ s(π)sπ» (π) + π(π)s(π)sπ» (π)

(19)

π(π)πβπ r(π)sπ» (π) + π(π)r(π)sπ» (π)

(20)

π=1

and Z(π) =

πβ1 β π=1

Eqation (18) becomes: H(π) = Z(π)πβ1 (π)

(21)

Isolating the term corresponding to π = π β 1 from the rest of the summation on the right-hand side of (19), we may write: ] [πβ2 β πβπ π» π» π(π) s(π)s (π) + π(π)s(π β 1)s (π β 1) π(π) = π=1

+ π(π)s(π)sπ» (π) (22)

(s,r)βπ

At time instant π, the constraint set πΆπ is defined as the set of all H(π) that satisfy (13) for the pilot-received signal pairs (s(π), r(π)): { } (15) πΆπ = H(π) β πΆ π : β£β£r(π) β H(π)s(π)β£β£ β€ πΎ .

The expression inside the brackets on the right-hand side of (22) equals π(πβ1) assuming the forgetting factor of the cost function is close to 1. Hence, we have the following recursion for updating the value of π(π):

The idea behind the SM channel estimation is that if the estimated channel at time instant π β 1 (H(π β 1)) lies outside the constraint set πΆπ , i.e. β£β£r(π) β H(π β 1)s(π)β£β£ > πΎ, we will make the estimated channel at the next time instant (H(π)) lie on the closest boundary of πΆπ . Otherwise, there is no need to compute and the consumption of power can be greatly saved.

Similarly, we may use (20) to derive the following recursion for updating Z(π):

π(π) = π(π β 1) + π(π)s(π)sπ» (π)

Z(π) = Z(π β 1) + π(π)r(π)sπ» (π)

The proposed BEACON channel estimation method can be considered as the following optimization problem: minimize

πβ1 β π=1

π(π)πβ1 (π β 1)s(π)sπ» (π)π(π)πβ1 (π β 1) 1 + π(π)sπ» (π)πβ1 (π β 1)s(π) (25)

For convenience of computation, let: P(π) = πβ1 (π)

π(π)

πβπ

β£β£r(π) β H(π)s(π)β£β£ 2

subject to β£β£r(π) β H(π)s(π)β£β£ = πΎ

2

(16)

(24)

Then, using the matrix inversion lemma [9], we obtain the following recursive equation for the inverse of π(π): πβ1 (π) = πβ1 (πβ1)β

A. Proposed BEACON Channel Estimation

(23)

and

2

142

k(π) =

sπ» (π)P(π β 1) 1 + π(π)sπ» (π)P(π β 1)s(π)

(26) (27)

TABLE I S UMMARY OF THE BEACON C HANNEL E STIMATION A LGORITHM

Therefore, we may rewrite (21) and (25) as: H(π) = Z(π)P(π)

(28)

P(π) = P(π β 1) β π(π)P(π β 1)s(π)k(π)

(29)

Initialize the algorithm by setting H(0) = 0 P(0) = I For each instant of time, π=1, 2, ..., compute { π(π) = ( r(π) β H(π)β 1)s(π) β£β£π(π)β£β£ 1 β 1 , if β£β£π(π)β£β£ > πΎ, πΊ(π) πΎ π(π) = 0, otherwise. where πΊ(π) = sπ» (π)P(π β 1)s(π)

Then we substitute (24) and (29) into (28) to obtain a recursive equation for updating the channel matrix H(π): H(π) = H(π β 1) β π(π)H(π β 1)s(π)k(π) + π(π)r(π)sπ» (π)P(π)

(30)

sπ» (π)P(πβ1)

k(π) = 1+π(π)πΊ(π) H(π) = H(π β 1) + π(π)π(π)k(π) P(π) = P(π β 1) β π(π)P(π β 1)s(π)k(π)

By rearranging (27) , we can get: k(π) = sπ» (π)P(π β 1) β π(π)sπ» (π)P(π β 1)s(π)k(π) = sπ» (π) [P(π β 1) β π(π)P(π β 1)s(π)k(π)]

B. BEACON Channel Estimation with a Time-Varying Bound

= sπ» (π)P(π) (31) Using (31) above, we get the desired recursive equation for updating the channel matrix H(π): H(π) = H(π β 1) β π(π)H(π β 1)s(π)k(π) + π(π)r(π)k(π) = H(π β 1) + π(π) [r(π) β H(π β 1)s(π)] k(π) = H(π β 1) + π(π)π(π)k(π) (32) where π(π) = r(π)βH(πβ1)s(π) denotes the prediction error vector at time instant π. Then we can get the error vector: e(π) = r(π) β H(π)s(π)

(33)

By substituting (32) into (33), we have: e(π) = r(π) β [H(π β 1) + π(π)π(π)k(π)] s(π) = r(π) β H(π β 1)s(π) β π(π)π(π)k(π)s(π) sπ» (π)P(π β 1)s(π) = π(π) β π(π)π(π) 1 + π(π)sπ» (π)P(π β 1)s(π) πΊ(π) (34) = π(π) β π(π)π(π) 1 + π(π)πΊ(π) [ ] π(π)πΊ(π) = π(π) 1 β 1 + π(π)πΊ(π) 1 = π(π) 1 + π(π)πΊ(π) where πΊ(π) = sH (π)P(π β 1)s(π). The constraint set is described as: 1 β£β£ β€ πΎ (35) β£β£e(π)β£β£ = β£β£π(π) 1 + π(π)πΊ(π) If β£β£π(π)β£β£ > πΎ, then the previous solution lies outside the constraint set. We can choose the constraint value β£β£e(π)β£β£ equal to πΎ so that the new solution lies on the closest boundary of the constraint set. Therefore: 1 = πΎ. (36) β£β£e(π)β£β£ = β£β£π(π)β£β£ β£1 + π(π)πΊ(π)β£ Hence the optimal forgetting factor at the πth iteration can be expressed as: ) ( β£β£π(π)β£β£ 1 β1 (37) π(π) = πΊ(π) πΎ Table I shows a summary of the BEACON channel estimation algorithm which will be used for the simulations.

In order to obtain the optimal error bound at each time instant, in this section we introduce an error bound function which can adjust the error bound automatically with the update of the channel estimate. It can be expressed as: β (38) πΎ(π + 1) = (1 β π½)πΎ(π) + π½ πΌβ£β£H(π)β£β£2 π 2 , where π½ is the forgetting factor, πΌ is the tuning parameter, and π 2 is the variance of the noise which is assumed to be known at the destinations. IV. S IMULATIONS In this section, we numerically study the performance of our proposed BEACON channel estimation method as well as the design of the optimal error bound. We consider a 3-hop wireless sensor network. The number of sources (ππ ), two groups of relays (ππ(1) , ππ(2) ) and destinations (ππ ) are 2, 4, 4, 3 respectively. We consider an AF cooperation protocol and the amplification coefficient of each relay is set to 1 for the purpose of simplification. We choose Hπ as our estimated channel because it is the most significant and most complex channel among all channels of the WSN system. During each phase, the sources and each group of relays transmit the QPSK modulated samples in the form of packets containing 2000 pilot symbols. The noise at the destinations is modeled as circularly symmetric complex Gaussian random variables with zero mean. The SNR is fixed at 10 dB. Fig. 3 shows the mean square error (MSE) performance of our proposed BEACON channel estimation, and compares it with conventional RLS channel estimation. Also, the minimum-mean-square error (MMSE) channel estimator which requires the full a priori knowledge of the channel correlation matrix and the noise variance is used here for reference. For the BEACON estimator, we choose four fixed error bounds (πΎ) ranging from 0.6 to 0.9. It can be seen that a higher value of πΎ results in worse MSE performance but a lower update rate (UR). It means the update is selective which can reduce the computational complexity and power consumption. In the case of an error bound equal to 0.6, the BEACON algorithm outperforms the conventional RLS algorithm (with a forgetting factor of 0.998) in terms of convergence speed and steady state with a slightly reduced UR (0.9128). When the error bound is increased to 0.8, although its convergence speed is slower than RLS channel estimation, the final MSE is comparable with a much lower UR (0.4356).

143

0

β2

10

10

RLS Ξ»=0.998 BEACON Ξ³=0.6 UR=0.9128 BEACON Ξ³=0.7 UR=0.7133 BEACON Ξ³=0.8 UR=0.4356 BEACON Ξ³=0.9 UR=0.2228 MMSE

β1

β3

10

β2

Mean Square Error

Mean Square Error

10

RLS Ξ»=0.998 BEACON Ξ³=0.6 BEACON Ξ³=0.7 BEACON Ξ³=0.8 BEACON Ξ³=0.9 BEACON with TVB Ξ±=3 Ξ²=0.001 UR=0.4003

10

β3

10

β4

10

β5

β4

10

10

β5

10

β6

10

0

500

1000 Received Symbols

1500

2000

0

2

4

6

8

10 SNR (dB)

12

14

16

18

20

Fig. 3. MSE performance of the BEACON channel estimation of Hπ compared with the RLS channel estimation.

Fig. 5. Channel estimation MSEs versus SNR for both the fixed bound and time-varying bound.

Fig. 4 illustrates the performance when we apply the timevarying bound (TVB) into the BEACON channel estimation. We set πΌ to 3 and π½ to 0.001. Our proposed algorithm can achieve very similar performance to the conventional RLS channel estimation with a substantial reduction in the UR. Therefore, the computational complexity is significantly reduced which is our goal.

V. C ONCLUSIONS We have proposed a BEACON channel estimation method based on a time-varying bound for cooperative wireless sensor networks. It has been shown that our proposed method can achieve similar performance to conventional RLS channel estimation, offering reduced computational complexity. Furthermore, the incorporation of the time-varying bound function makes it robust to changes in the environment. These features are desirable for WSNs and bring about a significant reduction in energy consumption.

0

10

RLS Ξ»=0.998 BEACON with TVB Ξ±=3 Ξ²=0.001 UR=0.4003 MMSE

β1

Mean Square Error

10

R EFERENCES

β2

10

β3

10

β4

10

β5

10

0

500

1000 Received Symbols

1500

2000

Fig. 4. MSE performance of the BEACON channel estimation with a timevarying bound.

In order to test our proposed channel estimation algorithm in a time-varying environment, the MSE versus SNR performance of the BEACON channel estimation methods is displayed with fixed error bounds and the proposed time-varying error bound in Fig. 5. In the cases of fixed error bounds, the MSE is lower bounded at different values for different error bounds. When the SNR is larger than a specified value, their MSEs will become worse. However, when the time-varying error bound is applied to the BEACON channel estimation, it can be observed that the MSE keeps on decreasing alone with the increase of the SNR. It shows the robustness to the time-varying environment of our proposed algorithm.

144

[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, βA Survey on Sensor Networks,β IEEE Communications Magazine, vol. 40, pp. 102-114, August 2002. [2] J. N. Laneman and G. W. Wornell, βCooperative diversity in wireless networks: Efficient protocols and outage behavior,β IEEE Transactions on Information Theory, vol. 50, no. 12, pp. 3062-3080, December 2004. [3] H. Li and P. D. Mitchell, βReservation packet medium access control for wireless sensor networks,β IEEE Personal, Indoor and Mobile Radio Communications Conference (PIMRC), 2008. [4] Y. W. Hong, W. J. Huang, F. H. chiu, and C. C. J. Kuo, βCooperative Communications in Resource-Constrained Wireless Networks,β IEEE Signal Processing Magazine, vol. 24, pp. 47-57, May 2007. [5] N. Khajehnouri and A.H. Sayed, βDistributed MMSE Relay Strategies for Wireless Sensor Networks,β IEEE Transactions on Signal Processing, vol. 55, no.7, pp. 3336-3348, July 2007. [6] R. Krishna, Z. Xiong, and S. Lambotharan, βA Cooperative MMSE Relay Strategy for Wireless Sensor Networks,β IEEE Signal Processing Letters, vol. 15, pp. 549-552, 2008. [7] Y. Li, B. Vucetic, Z. Zhou, and M. Dohler, βDistributed Adaptive Power Allocation for Wireless Relay Networks,β IEEE Transactions on Wireless Communications, vol. 6, no. 3, pp. 948-958, March 2007. [8] T. Wang, R. C. de Lamare and P. D. Mitchell, βLow-Complexity Channel Estimation for Cooperative Wireless Sensor Networks Based on Data Selection,β IEEE Vehicular Technology Conference (VTC), May 2010. [9] S. Haykin, Adaptive Filter Theory, 4th ed. Englewood Cliffs, NJ: Prentice-Hall, 2002. [10] R. C. de Lamare and P. S. R. Diniz, βSet-Membership Adaptive Algorithms Based on Time-Varying Error Bounds for CDMA Interference Suppression,β IEEE Transactions on Vehicular Technology, vol. 58, no. 2, pp. 644-654, February 2009. [11] S. Nagaraj, S. Gollamudi, S. Kapoor, and Y. F. Huang, βBEACON: An Adaptive Set-Membership Filtering Technique with Sparse Updatesβ IEEE Transactions on Signal Processing, vol. 47, pp. 2928-2941, Nov. 1999. [12] S. Dasgupta and Y. F. Huang, βAsymptotically Convergent Modified Recursive Least-Squares with Data Dependent Updating and Forgetting Factorβ 24th IEEE Conference on Decision and Control, pp. 10671071, 1985.