17-08-2013, 03:23 PM
Channel Estimation Modeling
Channel Estimation .pdf (Size: 45.88 KB / Downloads: 180)
Introduction
The radio channels in mobile radio systems are usually multipath fading channels, which are
causing intersymbol interference (ISI) in the received signal. To remove ISI from the signal, many
kind of equalizers can be used. Detection algorithms based on trellis search (like MLSE or MAP)
offer a good receiver performance, but still often not too much computation. Therefore, these
algorithms are currently quite popular.
However, these detectors require knowledge on the channel impulse response (CIR), which can
be provided by a separate channel estimator. Usually the channel estimation is based on the
known sequence of bits, which is unique for a certain transmitter and which is repeated in every
transmission burst. Thus, the channel estimator is able to estimate CIR for each burst separately
by exploiting the known transmitted bits and the corresponding received samples.
In this report we give first some general background information on channel estimation. Then
we introduce Least-squares (LS) channel estimation techniques. Normal LS channel estimation for
single signal is just from any textbook, but the chapter of joint channel estimation for 2 co-channels
simultaneously is based on several our own publications. Some comments on the simulation of
joint channel estimation system are given, too.
Background for channel estimation
Fig.1 shows a generic simulation layout for a TDMA based mobile system, which exploits
channel estimation and signal detection operations in equalisation. The digital source is usually
protected by channel coding and interleaved against fading phenomenon, after which the binary
signal is modulated and transmitted over multipath fading channel. Additive noise is added and the
sum signal is received.
Due to the multipath channel there is some intersymbol interference (ISI) in the received signal.
Therefore a signal detector (like MLSE or MAP) needs to know channel impulse response (CIR)
characteristics to ensure successful equalisation (removal of ISI). Note that equalization without
separate channel estimation (e.g., with linear, decision-feedback, blind equalizers [2]) is also
possible, but not discussed in this report. After detection the signal is deinterleaved and channel
decoded to extract the original message.
Channel estimator for single signal
Consider first a communication system, which is only corrupted by noise as depicted in Fig.3
below. Digital signal a is transmitted over a fading multipath channel hL, after which the signal has
memory of L symbols. Thermal noise is generated at the receiver and it is modelled by additive
white Gaussian noise n, which is sampled at the symbol rate. The demodulation problem here is to
detect the transmitted bits a from the received signal y. Besides the received signal the detector
ˆ , which are provided by a specific channel estimator device.
Simulation of joint channel estimation
Simulation layouts for joint channel estimation and single channel estimation are shown in Fig.5
and 6, respectively. In both cases there is similar co-channel interference present, but only joint
channel estimator takes it into account. In the latter case, the interference can be modelled by any
random binary signal, which is just modulated and transmitted over a multipath channel. But for
joint channel estimation it is required to send a proper training sequence also for the interfering
signal, hence the burst formatting is very important for the interferer also. Shortly, joint channel
estimation requires more accurate modeling for the interferer, because the receiver exploits some
known information on the interference as well.
Conclusions
This report presents some approaches, how to model channel estimation in simulations. First
we show a general simulation layout, which indicates that MLSE or MAP type of detection
algorithms require a separate channel estimator to provide CIR estimate. It is also shown that the
estimation is usually based on the known training bits and corresponding received samples.
LS channel estimation is thoroughly described. First we present the usual LS channel
estimation for a single signal in the presence of noise. Then we enlarge the estimation for 2 co-
channel signals simultaneously, which is needed by a specific joint detection algorithm. This joint
channel estimation requires a careful design of training sequences, since the cross-correlation
properties should also be good for the sequences. When this joint channel estimation is simulated,
one has to note that the interfering signal needs a proper modeling as well, because it is exploited
in the receiver. Normally, interference can be just modulated random binary signal without any
burst formatting.