values. In this section, we study the KF that is used
to predict variations of the channel in time domain
and which can also be applicable to frequency
domain.
The principle behind KF applied to channel
estimation is that we can represent the channel
frequency response, at time, as an infinite order
autoregressive (AR) process (Karakaya et al., 2009).
Figure 5: System using KF and block type pilot
arrangement.
Where, k and A are the order and the coefficient
of the AR process, respectively. V is a white
Gaussian noise with zero mean and variance σ
2
. For
the case of first order AR process
and .
In order to reduce the computational complexity,
only the first order AR process, i.e., is considered.
Therefore, we can represent the vector form of the
channel frequency response at time n as:
Then, the channel estimate can be obtained by a
set of recursions (Ling and Ting, 2006):
The received symbol at time can be expressed in the
form of a linear regression model.
4 PROPOSED CHANNEL
ESTIMATION SCHEME
In this section introduce and explain the proposed
method to predict the channel in high Doppler
spread.
Fig. 6 shows the New Channel Estimator's block
diagram that was designed considering the lattice-
type reference signals arrangement of 3GPP LTE (so
far most of the studies focus on block-type or comb
type pilot arrangements).
Figure 6: Block diagram of the new channel estimator.
After FFT, the reference signals or pilots of the
first OFDM symbols are extracted and we can
estimate the channel frequency response (CFR)
using a simple method, such as LS. Using the
recursions of KF, we can estimate the variation of
CFR for the later OFDM symbols containing
reference signals. Then, we transform the CFR into
CIR and eliminate the taps with index larger than L;
this way we eliminate the noise contained in those
taps. Finally, we transform the CIR to CFR and
estimate the channel for the rest of the subcarriers
using WF in time dimension.
5 SIMULATION RESULTS
In this section we show the performance analysis of
channel estimation methods. The simulations are
performed in MATLAB using the simulation
parameters of 3GPP LTE-Advanced shown in Table
5. The time variant channel is modelled according to
the values given for LTE extended channel models
in Table 1 and the maximum Doppler Frequency
values for each channel given in Table 3. The
frequency power spectrum follows the Jakes model.
Following the results of section II, we assume the
channel to be constant for 1 OFDM symbol.
Fig. 7 shows the BER performance of different
channel estimation methods in EPA channel with
max. Doppler Frequency of 5Hz. For this case, the
motion speed is low; therefore, as shown in sections
II, the channel suffers little variation in time within
one subframe and techniques like LS or MMSE with
linear interpolation produce good results.
Fig. 8 shows the BER performance of different
channel estimation methods in EVA channel with
max. Doppler frequency of 70Hz. In this case, the
mobile user moves with medium speed; therefore, as
shown in section II, the channel suffers more
variation in time within one subframe compared to
the case of EPA. We can observe that the BER
obtained with LS and MMSE starts to separate from
the actual value, but WF and New CE produce more
accurate results.
Fig. 9 shows the BER performance of different
channel estimation methods in ETU channel with
max. Doppler frequency of 300Hz. In this case, the
NovelChannelEstimationAlgorithmusingVariousFilterDesigninLTE-AdvancedSystem
123