Skip to main content

Optimal choice of test signals for linear channel estimation using second order statistics


A time-varying acoustic channel may be estimated by an appropriate inference using the output from a periodic test signal. In this paper it is shown how to do this in a way that takes full account of the past history of the background noise and the past history of the channel. An explicit formula is obtained for the optimal linear estimator that may be used for rapid channel estimation for a given test signal when we know the autocovariance or power spectrum of the interfering noise and the autocovariance of the echo channel variation.

Given this closed formula for the optimal estimator of the channel impulse response, an efficient method for determining the optimal test signal, subject to a constraint on the test signal power, given the history of the channel and the noise, is developed. We show that if the second order statistics of the channel or the noise are known, then the optimal test signal is not white. The method includes an explicit formula for the optimal test signal given a fixed estimator. A model of channel variation which is realistic while having less complexity than a full second-order statistical model, and therefore is more amenable to robust estimation, is used in the experiments which illustrate the performance of the optimal test signals and the channel estimation method. Matrix calculus identities required for the derivation of this expression for the optimal estimator are stated and proved in the Appendixes 1 and 2.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3


  • Benesty J (2006) Anonparametric VSS NLMS algorithm. IEEE Signal Process Lett 13(10):581–584

    Article  Google Scholar 

  • Birkett A, Goubran R (1995) Limitations of handsfree acoustic echo cancellers due to nonlinear loudspeaker distortion and enclosure vibration effects. In: The 1995 IEEE ASSP workshop on applications of signal processing to audio and acoustics, 15–18 October, pp 103–106

    Chapter  Google Scholar 

  • Boute R (1992) The Euclidean definition of the functions div and mod. ACM Trans Program Lang Syst 14(2):127–144

    Article  Google Scholar 

  • Cimini L (1999) Orthogonal frequency division multiplexing for wireless channels. Slide tutorial.

  • Costa M, Mermudez J (2008) A noise resilient variable step-size LMS algorithm. Signal Process 88(3):733–748

    Article  MATH  Google Scholar 

  • der Ouderaa EV, Schoukens J, Renneboog J (1988) Peak factor minimization of input and output signals of linear systems. IEEE Trans Instrum Meas 37(2):207–212

    Article  Google Scholar 

  • Edfors O, Sandell M, deBeek J, Wilson S, Borjesson P (1995) On channel estimation in OFDM systems. In: IEEE conference on vehicular technology, vol 2, pp 815–819

    Google Scholar 

  • Edfors O, Sandell M, deBeek J, Wilson S, Borjesson P (1996) OFDM channel estimation by singular value decomposition. In: 46th IEEE conference on vehicular technology “Mobile technology for the human race”, vol 2, pp 923–927

    Google Scholar 

  • Engels M (2002) Wireless OFDM systems: how to make them work. Kluwer Academic, Dordrecht

    Book  Google Scholar 

  • Felhauer T, Baier P, Konig W, Mohr W (1993) Optimum spread spectrum signals for wideband channel sounding. Electron Lett 29(6):563–564

    Article  Google Scholar 

  • Heinle F, Rabenstein R, Stenger A (1997) Measuring the linear and nonlinear properties of electro-acoustic transmission systems. In: International workshop on acoustic echo and noise control, pp 33–36

    Google Scholar 

  • Kates J (1991) Feedback cancellation in hearing aids: results from a computer simulation. IEEE Trans Signal Process 39(3):553–562

    Article  Google Scholar 

  • Li Y (2000) Pilot-symbol-aided channel estimation for OFDM in wireless systems. IEEE Trans Veh Technol 49(4):1207–1215

    Article  Google Scholar 

  • Li Y, Stuber G (2006) Orthogonal frequency division multiplexing for wireless communications. Springer, Berlin. Chap 5

    Book  Google Scholar 

  • Luenberger D (1969) Optimisation by vector space methods. Wiley, New York. Chap 9

    Google Scholar 

  • Mader A, Puder H, Schmidt G (2000) Step-size control for acoustic echo cancellation filters an overview. Signal Process 80(8):1697–1719

    Article  MATH  Google Scholar 

  • MathWorks (2008) fmincon.

  • Molina A, Fannin P, Timoney J (1995) Generation of optimum excitation waveforms for mobile radio channel sounding. IEEE Trans Veh Technol 44(2):275–279

    Article  Google Scholar 

  • Negi R, Cioffi J (1998) Pilot tone selection for channel estimation in a mobile OFDM system. IEEE Trans Consum Electron 44(3)

  • Noh M, Lee Y, Park H (2006) Low complexity LMMSE channel estimation for OFDM. IEE Proc, Commun 153(5):645–650

    Article  Google Scholar 

  • Petersen K, Pederson M (2008) The matrix cookbook. Tech. rep.

  • Schafhuber D, Matz G, Hlawatsch F (2003) Adaptive wiener filters for time-varying channel estimation in wireless OFDM systems. In: Proceedings of the IEEE international conference on acoustics, speech and signal processing (ICASSP ’03), 6–10 April, Hong Kong, vol 4, pp 688–691.

    Google Scholar 

  • Schroeder M (1970) Synthesis of low-peak-factor signals and binary sequences with low autocorrelation. IEEE Trans Inf Theory 16(1):85–89

    Article  Google Scholar 

  • Shen Y, Martinez E (2006) Channel estimation in OFDM systems. Application note, Freescale semiconductor

  • Sklar B (1997) Rayleigh fading channels in mobile digital communication systems. Part I. Characterization. IEEE Commun Mag 35(7):90–100

    Article  Google Scholar 

  • Srivastava V, Ho CK, Sun S, (2004) Robust MMSE channel estimation in OFDM systems with practical timing synchronization. In: Wireless communications and networking conference. 2004 (WCNC 2004) IEEE, pp 711–716

    Chapter  Google Scholar 

  • Stapleton J (1995) Linear statistical models (Chap 3). Wiley, New York

    Book  Google Scholar 

  • Van den Bos A (1987) A new method for synthesis of low-peak-factor signals. IEEE Trans Acoust Speech Signal Process 35(1):120–122

    Article  MathSciNet  Google Scholar 

  • van Houwelingen J (2001) Shrinkage and penalized likelihood as methods to improve predictive accuracy. Stat Neerl 55(1):17–34

    Article  MATH  Google Scholar 

  • Wei WY, Gelfand S, Krogmeier J (2001) Noise-constrained least mean squares algorithm. IEEE Trans Signal Process 49(9):1961–1970

    Article  Google Scholar 

Download references

Author information

Authors and Affiliations


Corresponding author

Correspondence to Stephen Braithwaite.


Appendix 1: Some identities of traces and their derivatives with proofs

5.1 Identity 1.1

Let A be an arbitrary real matrix. Let x be an arbitrary real vector, and \(X = \operatorname{Circ}(x)\). Then

$$\frac{\partial}{\partial x} \operatorname{tr}(XA) = \operatorname{traces}(A) $$



5.2 Identity 1.2

Let A be an arbitrary real matrix. Let x be an arbitrary real vector, and \(X = \operatorname{Circ}(x)\). Then

$$\frac{\partial}{\partial x} \operatorname{tr}\bigl(X^T A\bigr) = \operatorname{traces} \bigl(A^T\bigr) $$



5.3 Identity 1.3

Let A and B be arbitrary real matrices. Let x be an arbitrary real vector, and \(X = \operatorname{Circ}(x)\). Then

$$\frac{\partial}{\partial x} \operatorname{tr}\bigl(X^T A X B\bigr) = \operatorname{traces} \bigl(B^T X^T A^T\bigr) + \operatorname{traces}\bigl(B X^T A\bigr) $$


$$\operatorname{tr}\bigl(X^T A X B\bigr) = \operatorname{tr}\bigl(X^T (A X B)\bigr) = \operatorname{tr}\bigl(X \bigl(B X^T A\bigr) \bigr) $$

Therefore, applying Identity 1.1 and Identity 1.2 using the product rule:-

$$\frac{\partial}{\partial x} \operatorname{tr}\bigl(X^T A X B\bigr) = \operatorname{traces} \bigl(B^T X^T A^T\bigr) + \operatorname{traces}\bigl(B X^T A\bigr) $$


5.4 Identity 1.4

Let A and B be a arbitrary real matrices. Let x be an arbitrary real vector, and \(X = \operatorname{Circ}(x)\). Then

$$\operatorname{traces}\bigl(A X^T B\bigr) = G x $$

where the elements of the matrix G are given by

$$G_{s,k} = \sum_{i=0}^{n-1} \sum _{j=0}^{n-1} A_{i,j} B_{(j+k)\% n,(i+s)\%n} $$


We can define \(\operatorname{traces}()\) by

$$\operatorname{traces}(P)_{s} = \sum_{i=0}^{n-1} P_{i,(i+s)\%n} $$

for each index s. Therefore:-


5.5 Identity 1.5

Let A and B be a arbitrary real matrices and d be an arbitrary real vector. Then

$$\operatorname{diag}\bigl(A \operatorname{Diag}(d) B\bigr) = \bigl(A \diamond B^T \bigr)d $$


Let \(D = \operatorname{Diag}(d)\)


Appendix 2: Some further identities

6.1 Identity 2.1

Given a vector of random variables x and symmetric matrix A, then

$$E\bigl( x^{T} A x \bigr) = \operatorname{tr}(A C_x) + m^T A m $$

Where C x is the covariance matrix of x and m is the mean of x (Petersen and Pederson 2008; Identity 306).

6.2 Identity 2.2

Given arbitrary square matrices A, B and X, then (Petersen and Pederson 2008; Identity 93)

$$\frac{\partial}{\partial X} \operatorname{tr}(A X B) = A^T B^T. $$

6.3 Identity 2.3

Given arbitrary square matrices A and X, then (Petersen and Pederson 2008; Identity 101)

$$\frac{\partial}{\partial X} \operatorname{tr}\bigl(X A X^T\bigr) = X A^T + X A. $$

6.4 Identity 2.4

Given a square matrix Y that is a function of a scalar x, then (Petersen and Pederson 2008; Identity 53)

$$\frac{\partial}{\partial x} \bigl(Y^{-1}\bigr) = -Y^{-1} \frac{\partial Y}{\partial x}Y^{-1}. $$

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Braithwaite, S., Addie, R.G. Optimal choice of test signals for linear channel estimation using second order statistics. Optim Eng 15, 93–118 (2014).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


  • Acoustic
  • Echo
  • Autocovariance
  • Channel estimation
  • Matlab