## Abstract

This paper investigates the parameter-based (PB) channel estimation for orthogonal frequency division multiplexing over time-varying channels. The time-varying channel impulse response (CIR) can be represented using limited number of channel parameters. Instead of estimating the CIR directly, the PB approach estimates these parameters, from which, the CIR can be regenerated. To estimate these parameters, a differential scheme, utilizing the repetitive structure of the training signal, is proposed in this paper. When expressing the received signal in time–frequency-representation form, the repetitive structure can cause a phase difference between consecutive signal branches. This provides the chance for parameter estimation. Space domain sampling using multiple receive antennas are also employed to gain unique solution for parameter estimation. The effectiveness of the proposed algorithm is demonstrated by computer simulations.

### Similar content being viewed by others

## References

Bingham, J. A. C. (1990). Multicarrier modulation for data transmission: An idea whose time has come.

*IEEE Communications Magazine,**37*, 5–14.May, T., Rohling, H., & Engels, V. (1998). Performance analysis of Viterbi decoding for 64-DAPSK and 64-QAM modulated signal.

*IEEE Transactions on Communications,**46*, 182–190.Visintin, M. (1996). Karhunen–Loeve expansion of a fast Rayleigh fading process.

*IEEE Electronics Letters,**32*(8), 1712–1713.Teo, K. D., & Ohno, S. (2005). Optimal MMSE finite parameter model for doubly-selective channels. In

*Proceedings of IEEE Global Telecommunication Conference (GLOBECOME)*(pp. 3503–3507).Zemen, T., & Mechlenbrauker, C. F. (2005). Time-varaint channel estimation using discrete prolate spheroidal sequences.

*IEEE Transactions on Signal Processing,**53*(9), 3597–3607.Ma, X., Giannakis, G., & Ohno, S. (2003). Optimal training for block transmissions over doubly-selective fading channels.

*IEEE Transactions on Signal Processing,**51*(5), 1351–1366.Tang, Z., Cannizzaro, R. C., Leus, G., & Banelli, P. (2007). Pilot-assisted time-varying channel estimation for OFDM systems.

*IEEE Transactions on Signal Processing,**55*(5), 2226–2238.Sayeed, A. M., & Aazhang, B. (1999). Joint multipath-Dopper diversity in mobile wireless communications.

*IEEE Transactions on Communications,**47*(1), 123–132.Kannu, A. P., Schniter, P. (2005). MSE-optimal training for linear time-varying channels. In

*Proceedings of IEEE international conference acoustics speech, signal process. (ICASSP)*(pp. iii/789–iii/792).Tomasin, S., Gorokhov, A., Yang, H., & Linnartz, J. P. (2005). Iterative interference cancellation and channel estimation for mobile OFDM.

*IEEE Transactions on Wireless Communications,**4*(1), 238–245.Halwatsch, F., & Matz, G. (2011).

*Wireless communications over rapidly time-varying channels*. Waltham: Academic Press of Elsevier. 2011.Du, Z., Song, X., Cheng, J., & Beaulieu, N. C. (2009).

*Aconergence study of iterative channel estimation algorithms for OFDM systems in dispersive time-varying channels*. WCNC.Du, Z., Song, X., Cheng, J., & Beaulieu, N. C. (2011). Maximum likelihood based channel estimation for macrocellular OFDM uplinks in dispersive time-varying channels.

*IEEE Transactions on Wireless Communications,**10*(1), 176–187.Liu, Y., Tan, Z., Wang, H., Xu, S., & Kwak, K. S. (2012). Channel estimation for macrocellular OFDM uplinks in time-varying channels.

*IEEE Transactions on Vehicular Technology,**61*(4), 1709–1717.Goldsmith, G. (2005).

*Wireless communications*. Cambridge: Cambridge University Press.Giannakis, G. B., & Tepedelenlioglu, C. (1998). Basis expansion models and diversity techniques for blind identificatino and equalizatino of time-varying channels.

*Proceedings of the IEEE,**86*(10), 1969–1986.Clarke, R. H. (1968). A statistical theory of mobile-radio reception.

*Bell System Technical Journal,**47*(6), 957–1000.Kay, S. M. (1993).

*Fundamentals of statistical signal prcoessing: estimation theory*. Englewood Cliffs, NJ: Prentice-Hall.Morelli, M., & Mengali, U. (1999). An improved frequency offset estimator for OFDM application.

*IEEE Communications Letters,**3*(3), 75–77.Meyer, C. D. (2004).

*Matrix analysis and applied linear algebra*(Vol. 1). Cambridge: Cambridge University Press.Rugini, L., Banelli, P., & Leus, G. (2005). Block DFE and windowing for Doppler-affected OFDM systems. In

*IEEE signal process. Workshop signal process. Advance Wireless Communications (SPAWC)*, pp. 470–474.Pop, M. F., & Beaulieu, N. C. (2001). Limitations of sum-of-sinusoids fading channel simulators.

*IEEE Transactions on Communications,**48*, 699–708.Xiao, C., Zheng, Y. R., & Beaulieu, N. C. (2006). Novel sum-of-sinusoids simulation models for Rayleigh and Rician fading channels.

*IEEE Transaction of Wireless Communication,**5*(12), 3667–3679.

## Acknowledgments

This work was supported by the National Natural Science Foundation of China (Grant No. 61302106 and No. 61202079), Natural Science Foundation of Hebei Province (No. F2014502029) and the Fundamental Research Funds for the Central Universities (No. 2014MS100).

## Author information

### Authors and Affiliations

### Corresponding author

## Appendix

### Appendix

From the structure of \( \varvec{\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle-}$}}{C} } \), to prove matrix \( \varvec{\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle-}$}}{C} } \) is a column full rank matrix, we first assume that the estimation of Doppler frequency is ideal.

To prove ** C** is a column full rank matrix, we assume there are a set of coefficients

*α*

_{ l,q }that can make

where *λ*
^{r}_{
qL+1}
is the (*qL* + 1)th element of \( {\varvec{\Lambda}}^{\left( r \right)} \), \( \varvec{s}_{l} = \varvec{F}^{H} t_{l} \) and *t*
_{
l
} is the *l*th column of product matrix \( \varvec{TF}_{L} \). Actually, we have \( \tfrac{Q + 1}{2} \) such equations with *r* changing from 0 to \( \tfrac{Q - 1}{2} \). Multiplying \( \varvec{s}_{{l_{0} }}^{H} \) on both sided of (41) with where 0 ≤ *l*
_{0} ≤ *L* − 1, we have

Note that

where \( \odot \) denotes the circular convolution, \( \varvec{\omega}_{F} \left( q \right) = \varvec{F}^{H}\varvec{\omega}\left( q \right) \) and \( \varvec{\omega}\left( q \right) \) is the vector representation of elements on the diagonal line of \( {\varvec{\Omega}}\left( {v_{q} } \right) \), that is \( \varvec{\omega}\left( q \right) = {\varvec{\Omega}}\left( {v_{q} } \right){\mathbf{I}}_{N} \) with \( {\mathbf{I}}_{N} \) being an *N* × 1 all ones vector. The third equation in (43) is due to the fact that the circular convolution can be represented as the form of matrix product, and \( \varvec{t}_{l}^{\left( s \right)} \) denotes the circular shift of \( \varvec{t}_{l}^{{}} \). By multiplying \( \varvec{s}_{{l_{0} }}^{H} \) on both sides of (43), we can obtain

Recall the fact that the frequency domain training sequence is obtained by interleaving the m-sequence with null subcarriers. Hence, we have \( \varvec{t}_{{l_{0} }}^{H} \varvec{t}_{l}^{(s)} = 0 \) when *s* is odd. When *s* is even, by noting that \( \varvec{t}_{l}^{{}} \) is the m-sequence modulated by a carrier \( e^{{ - j\tfrac{2\pi ln}{N}}} \), from the derivation in [8], it is easy to find that \( \varvec{t}_{{l_{0} }}^{H} \varvec{t}_{l}^{(s)} = N\sigma_{t}^{2} \) only when *l* = *l*
_{0}, *s* = 0, and \( \varvec{t}_{{l_{0} }}^{H} \varvec{t}_{l}^{(s)} = 0 \) for other *l*’s and *s*’s. In summary, we can derive the following the relation

With the relation (45), (44) can be simplified as

where \( \left[ {\varvec{\omega}_{F} \left( q \right)} \right]_{(0)} \) denotes the 0th element of vector \( \varvec{\omega}_{F} \left( q \right) \), that is, the d.c. component of \( \varvec{\omega}_{F} \left( q \right) \). For large enough *N*, it is easy to derive an approximation of \( \left[ {\varvec{\omega}_{F} \left( q \right)} \right]_{(0)} \) as

If denoting \( \left[ {\varvec{\omega}_{F} \left( q \right)} \right]_{(0)} \lambda_{qL + l}^{r} = m(r,q) \), (42) can be simplified as

When *r* changes from 0 to \( \tfrac{Q - 1}{2} \), we can obtain \( \tfrac{Q + 1}{2} \) such equations, and the matrix form of those equations can be given as

where \( \left[ \varvec{M} \right]_{r,q} = m(r,q) \) and \( \varvec{\alpha}_{{l_{0} }} = \left( {\alpha_{{l_{0} ,0}} ,\alpha_{{l_{0} ,1}} , \ldots ,\alpha_{{l_{0} ,\tfrac{Q - 1}{2}}} } \right)^{T} \). For given system parameters, e.g. *d*
_{0} = 0.35 and *Q* = 9, ** M** becomes a deterministic matrix, which depends only on maximum Doppler frequency V. It is easy to verify that matrix

**is invertable when 0 ≤**

*M**V*< 1. Therefore, there is only zero vector as the solution for (49), that is, \( \alpha_{{l_{0} ,q}} = 0 \) for \( q = 0,1, \ldots ,\tfrac{Q - 1}{2} \). Similar conclusion can be obtained for other

*l*

_{0}∊ [0,

*L*− 1]. This means that only all zeros coefficients can satisfy the relation in (41). Hence, \( \varvec{C} \) is a column full rank matrix.

In above, we assumed the Doppler frequency estimation is ideal. In the presence of Doppler frequency estimation error, the noise causes some perturbation \( \Delta \varvec{M} \) to (49) and thus we obtain

However, as shown in [20], small perturbation cannot reduce the matrix rank, that is

where \( r( \cdot ) \) denotes the matrix rank. Since \( \varvec{M} \) is a full rank matrix, \( \varvec{M + }\Delta \varvec{M} \) is also a full rank matrix and thus invertable. This makes \( \varvec{\alpha}_{{l_{0} }} = {\mathbf{0}} \) become the unique solution of (50). Correspondingly, matrix \( \varvec{\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle-}$}}{C} } \) is column full rank matrix as well.

## Rights and permissions

## About this article

### Cite this article

Han, D., Liu, Y., Ni, J. *et al.* Parameter Based Channel Estimation for OFDM Systems Over Time-Varying Channels.
*Wireless Pers Commun* **83**, 703–720 (2015). https://doi.org/10.1007/s11277-015-2418-y

Published:

Issue Date:

DOI: https://doi.org/10.1007/s11277-015-2418-y