Sorry, I have to add some information.
The simulation is implemented in Matlab. Both transmitter and receiver
RRCs are done in time domain. Its oversampling Fs is two times of
symbol rate Fd. The symbol is 16 QAM. The dispersive channel is a FIR
model who has 12 items in time span. The channel matching is done in
Thanks in advance.
Reply by ●January 8, 20052005-01-08
I am simulating a single carrier system in which signals are equalized
in frequency domain. The concept is similar to OFDM, except both FFT
and IFFT are reside in receiver side. Prefix is also used to deal with
dispersive channel response.
I have implemented the rooted raised cosine filter(Alpha=0.2) in both
transmitter and receiver. A static dispersive fading channel is used in
each frame. There is no problem with no fading channel, i.e. AWGN
channel. The first step I adopt is a simple zero forcing equalizer.
The input sampling frequency is two times symbol rate to comply with
the bandlimited width channel. FFT is then used to transform signal to
frequency domain from time domain. After matched filter processing in
frequency domain(including RRC and ideal channel matching), a sampling
rate reduction(2 to 1) is used to reduce the date rate to that of the
symbol rate. This has considered the aliasing combining effects.
As a test procedure of the equalizer, I don't add noise to the system.
So, I expect I could get no BER even I use a simple ZF equalizer. Now,
the equalizer seems not very well. At the beginning and the end
edges(It is about 4 to 8 symbols each end.) of a data block(The total
block length is 256), there are some slight signal distortion. In the
center part of the data block, signal is very good. Of course, this is
after IFFT transformation.
The dispersive channel is about 6 symbols long. The RRC tail is about 4
symbols long. I first used the prefix as long as 16 symbols. In order
to isolate problem, I have increased the prefix length to 32, even 64.
The problem still exists. This is really strange. From the equalized
signal waveform. It looks like the initial signal level related. There
are some low frequency modulation for the 4 to 8 symbols, i.e. a low
frequency but a noticable level fluctuation. I know here it is a signal
block based DSP. It is different from the normal time domain
processing. What would cause the edges signal worse? The prefix should
convert linear convolution to circular convolution. This is like an
ideal periodic signal. It shouldn't make the edges differnt from other
data. I am sure it is the edges resulting the average error rate about
10^(-3), not 0. But for an ideal circular convolution, this shouldn't
Could you give me some suggestion?
Any comments are highly appreciated.