Forums

relation of equalizer and channel coding

Started by buaawh June 4, 2008
Hello,

Well known that decision-feedback equalizer is generally superior than
linear equalizer, I was surprised to read the statement in a paper:

"Finally it is to mention that a combination of channel coding (Viterbi
Decoding) and DFE did not lead to a performance gain compared to channel
coding and linear equalization. It is well known that this combination
represents a critical one..."

what does "critical one" mean? Is there any simple interpretation about
why "DFE+channel coding"has no performance gain than "LE+annel coding"?

Thanks

buaawh wrote:
> Hello, > > Well known that decision-feedback equalizer is generally superior than > linear equalizer,
"It is well known that a motorcycle is generally superior to a refrigerator" Eh?
> I was surprised to read the statement in a paper: > > "Finally it is to mention that a combination of channel coding (Viterbi > Decoding) and DFE did not lead to a performance gain compared to channel > coding and linear equalization. It is well known that this combination > represents a critical one..."
It depends.
> what does "critical one" mean?
The meaning is "it depends".
> Is there any simple interpretation about > why "DFE+channel coding"has no performance gain than "LE+annel coding"?
DFE compensates zeroes with poles. LE compensates poles with zeroes. Both don't work well if used vice versa. Both are suboptimal. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
> > >buaawh wrote: >> Hello, >> >> Well known that decision-feedback equalizer is generally superior than >> linear equalizer, > >"It is well known that a motorcycle is generally superior to a >refrigerator" > >Eh? > > >> I was surprised to read the statement in a paper: >> >> "Finally it is to mention that a combination of channel coding
(Viterbi
>> Decoding) and DFE did not lead to a performance gain compared to
channel
>> coding and linear equalization. It is well known that this combination >> represents a critical one..." > >It depends. > >> what does "critical one" mean? > >The meaning is "it depends". > >> Is there any simple interpretation about >> why "DFE+channel coding"has no performance gain than "LE+annel
coding"?
> >DFE compensates zeroes with poles. >LE compensates poles with zeroes. >Both don't work well if used vice versa. >Both are suboptimal. > > >Vladimir Vassilevsky >DSP and Mixed Signal Design Consultant >http://www.abvolt.com >
%%%% That statement is correct. I realised this fact when I did the simulation for DFE+coding and when I got strange results, I searched for the reasos. The reason behind this is the fact that DFE suffers from error propagation. If error was made during ahrd decision it will propagate thru DFE and will creat 'burst of error', which can not be corrected by any error correcting, not even BCJR. You can achive little bit of coding gain, if you use interleaver at transmitter and deinterleaver at receiver. Now, this littlebit of also 'depends' on type of channel. Definately u will get gain, no matter how worst the channel is, but this gain will be at high SNR. The reason that I have given here can be found in a very good 'Wireless Communications' BY Andrea Goldsmith, Cambridge University press. And to run any simulation, use famouse channels given by Proakis. I generally use channel B with taps channel=[0.407 0.815 0.407] Hope this helps. Chintan
>> >> >>buaawh wrote: >>> Hello, >>> >>> Well known that decision-feedback equalizer is generally superior
than
>>> linear equalizer, >> >>"It is well known that a motorcycle is generally superior to a >>refrigerator" >> >>Eh? >> >> >>> I was surprised to read the statement in a paper: >>> >>> "Finally it is to mention that a combination of channel coding >(Viterbi >>> Decoding) and DFE did not lead to a performance gain compared to >channel >>> coding and linear equalization. It is well known that this
combination
>>> represents a critical one..." >> >>It depends. >> >>> what does "critical one" mean? >> >>The meaning is "it depends". >> >>> Is there any simple interpretation about >>> why "DFE+channel coding"has no performance gain than "LE+annel >coding"? >> >>DFE compensates zeroes with poles. >>LE compensates poles with zeroes. >>Both don't work well if used vice versa. >>Both are suboptimal. >> >> >>Vladimir Vassilevsky >>DSP and Mixed Signal Design Consultant >>http://www.abvolt.com >> >%%%% > >That statement is correct. I realised this fact when I did the
simulation
>for DFE+coding and when I got strange results, I searched for the
reasos.
> >The reason behind this is the fact that DFE suffers from error >propagation. If error was made during ahrd decision it will propagate
thru
>DFE and will creat 'burst of error', which can not be corrected by any >error correcting, not even BCJR. > >You can achive little bit of coding gain, if you use interleaver at >transmitter and deinterleaver at receiver. Now, this littlebit of also >'depends' on type of channel. Definately u will get gain, no matter how >worst the channel is, but this gain will be at high SNR. > >The reason that I have given here can be found in a very good 'Wireless >Communications' BY Andrea Goldsmith, Cambridge University press. > >And to run any simulation, use famouse channels given by Proakis. I >generally use channel B with taps channel=[0.407 0.815 0.407] > >Hope this helps. > >Chintan >
I get it. Also I learn that predictive DFE is suitble for coding system form Proakis's "Digital Communications". Much thanks!