Channel Estimation for OFDM System with Two Training Symbols Aided and Polynomial Fitting



  4_Channel Estimation for OFDM System.pdf (Size: 406.75 KB / Downloads: 13)

INTRODUCTION

ORTHOGONAL frequency division multiplexing
(OFDM) is a high spectral efficiency transmission
scheme. It divides a wideband signal into many orthogonal
subcarriers and induces a symbol period that is much
longer than the bit duration. However, a long symbol
period is sensitive to the time-varying channel in wireless
communications. The channel response might change during
an OFDM symbol period in a high-mobility environment.
Therefore, orthogonality among subcarriers is destroyed, and
inter-carrier interference (ICI) occurs [1]-[2].



CHANNEL ESTIMATION WITH TWO TRAINING
SYMBOLS AND POLYNOMIAL FITTING

As mentioned above, H can be transformed from the timedomain
matrix g. Because we do not exactly know the number
of paths, the worst case that the channel delay spread becomes
equal to the length of CP should be considered. If we estimate
the time domain matrix g rather than H, then the number of
unknowns can be reduced to NL. It has been shown that when
the channel response does not vary too fast, such as when
the OFDM symbol period is 8% of the channel coherence
time or less [11], the channel response varies almost linearly
within one OFDM symbol. If the channel response of each
path is modeled by a linear function, then only two unknowns
for each path need to be solved.


PERFORMANCE VERIFICATION

the performance of the presented channel
estimation method under the COST207 bad urban area (BU)
multipath fading channel [9]. A second-order polynomial is
used in the simulation. Two normalized Doppler frequencies
under various data symbols are tested. The number of subcarriers
is 512, and the CP length is 64. The training symbols are
pseudo-random noise (PN) sequences with quadrature phase
shift keying (QPSK) modulation, which is also employed for
all data symbols in the simulations.


CONCLUSION
A channel estimation method for OFDM systems in fast
time-varying channels has been presented. The presented
method uses the structure of two training symbols in combination
with polynomial fitting. Thus, it can provide enough
channel estimation accuracy for even a normalized Doppler
frequency as large as 0.1. The major advantage of using the
two training symbols is that no ICI can disturb the estimation.
The disadvantages of the presented channel method include:
(1) increased computational complexity, and (2) requirement
of a long buffer with +2 symbols. Furthermore, as the
variation cycle of each path response is inversely proportional
to the normalized Doppler frequency, the two training symbols
and the data symbols are better located within one-half of
the variation cycle.