Forums

Ungerboeck metric for MLSE Equalizer

Started by Pave...@gmail.com January 7, 2008
Hello.
Who can give me some description about Ungerboeck MLSE equalizer? I
use it for GSM burst equalization and get strange result - it works
bad. BER is increased as comparision with no equalization. I estimate
Channel Impulse Response and use it for signal reconstruction in
eaualizer but it works bad. So I make symbol-by-symbol decisions. Can
you guide me to additional information about MLSE equalization in GSM?
On Jan 7, 12:18 pm, "Pavel.Schu...@gmail.com"
<Pavel.Schu...@gmail.com> wrote:
> Hello. > Who can give me some description about Ungerboeck MLSE equalizer? I > use it for GSM burst equalization and get strange result - it works > bad. BER is increased as comparision with no equalization. I estimate > Channel Impulse Response and use it for signal reconstruction in > eaualizer but it works bad. So I make symbol-by-symbol decisions. Can > you guide me to additional information about MLSE equalization in GSM?
Hi Pavel, When you state that you are making "symbol-by-symbol decisions", do you mean that you're using this as a fall-back because your MLSE algo, which should be looking at symbol history, is not working? How many symbol periods does your MLSE algo span? Darol Klawetter
On Jan 7, 10:20&#2013266080;pm, Darol Klawetter <darol.klawet...@l-3com.com>
wrote:
> On Jan 7, 12:18 pm, "Pavel.Schu...@gmail.com" > > <Pavel.Schu...@gmail.com> wrote: > > Hello. > > Who can give me some description about Ungerboeck MLSE equalizer? I > > use it for GSM burst equalization and get strange result - it works > > bad. BER is increased as comparision with no equalization. I estimate > > Channel Impulse Response and use it for signal reconstruction in > > eaualizer but it works bad. So I make symbol-by-symbol decisions. Can > > you guide me to additional information about MLSE equalization in GSM? > > Hi Pavel, > > When you state that you are making "symbol-by-symbol decisions", do > you mean that you're using this as a fall-back because your MLSE algo, > which should be looking at symbol history, is not working? > > How many symbol periods does your MLSE algo span? > > Darol Klawetter
I mean that i conclude not to use equalizer at all. MLSE span is 5 symbol. Estimated channel impulse response is of 5 taps. And i use common Viterbi algorithm for detect most likelihood sequence. I think there is a bug in Viterbi algorithm's code. Or it can be due to metric computation mistake.
On Jan 7, 1:55 pm, "Pavel.Schu...@gmail.com" <Pavel.Schu...@gmail.com>
wrote:
> On Jan 7, 10:20 pm, Darol Klawetter <darol.klawet...@l-3com.com> > wrote: > > > > > On Jan 7, 12:18 pm, "Pavel.Schu...@gmail.com" > > > <Pavel.Schu...@gmail.com> wrote: > > > Hello. > > > Who can give me some description about Ungerboeck MLSE equalizer? I > > > use it for GSM burst equalization and get strange result - it works > > > bad. BER is increased as comparision with no equalization. I estimate > > > Channel Impulse Response and use it for signal reconstruction in > > > eaualizer but it works bad. So I make symbol-by-symbol decisions. Can > > > you guide me to additional information about MLSE equalization in GSM? > > > Hi Pavel, > > > When you state that you are making "symbol-by-symbol decisions", do > > you mean that you're using this as a fall-back because your MLSE algo, > > which should be looking at symbol history, is not working? > > > How many symbol periods does your MLSE algo span? > > > Darol Klawetter > > I mean that i conclude not to use equalizer at all. > MLSE span is 5 symbol. Estimated channel impulse response is of 5 > taps. And i use common Viterbi algorithm for detect most likelihood > sequence. I think there is a bug in Viterbi algorithm's code. Or it > can be due to metric computation mistake.
You may want to verify that your symbol timing recovery is adequate for the MLSE algo. The MLSE approach is sensitive to timing errors. In the past, I've chosen to fight ISI using a fractional DFE rather than MLSE because of this sensitivity, as well as the greater complexity of a MLSE approach.