Hi All, I am trying to figure out the *real* motivation behind using MLSE (Maximum Likelihood Sequence Detection) as opposed to a symbol-by-symbol detection that maximizes the log-likelihood function. As a tangible example, the standard optimum receiver for CPM (Continuous-Phase Modulation) signals is defined to be a correlator followed by an MLSE detector (Ref. Digital Communications by J. G. Proakis). My question, rephrased in the context of this example, is as follows: What difference would it make (in terms of error performance, complexity etc etc) if, instead of using the standard approach, I make decision on a symbol-by-symbol basis by correlating the received signal during each signaling interval with all possible waveforms that could have been transmitted, and picking the one that yields the largest correlation? I hope someone is able to respond to my question this time. My last one didn't elicit a single response :( Thanks for reading!
MLSE vs. Symbol-by-Symbol detection
Started by ●April 7, 2009
Reply by ●April 7, 20092009-04-07
cyclotron wrote:> Hi All, > > I am trying to figure out the *real* motivation behind using MLSE (Maximum > Likelihood Sequence Detection) as opposed to a symbol-by-symbol detection > that maximizes the log-likelihood function.The real motivation for MLSE is making your life miserable.> As a tangible example, the standard optimum receiver for CPM > (Continuous-Phase Modulation) signals is defined to be a correlator > followed by an MLSE detector (Ref. Digital Communications by J. G. > Proakis). My question, rephrased in the context of this example, is as > follows: > > What difference would it make (in terms of error performance, complexity > etc etc) if, instead of using the standard approach, I make decision on a > symbol-by-symbol basis by correlating the received signal during each > signaling interval with all possible waveforms that could have been > transmitted, and picking the one that yields the largest correlation?That entirely depends on the channel, the signal and the error rate. For the AWGN, uncoded GMSK BT ~ 0.5 and the error rate ~1e-3, the difference between MLSE and symbol by symbol detection is ~0.5dB. For the multipath channels, coded multilevel signals and low BER, the difference can be as much as 10dB.> I hope someone is able to respond to my question this time. My last one > didn't elicit a single response :(Because you should do your classwork yourself. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●April 7, 20092009-04-07
> > >cyclotron wrote: > >> Hi All, >> >> I am trying to figure out the *real* motivation behind using MLSE(Maximum>> Likelihood Sequence Detection) as opposed to a symbol-by-symboldetection>> that maximizes the log-likelihood function. > >The real motivation for MLSE is making your life miserable. > >> As a tangible example, the standard optimum receiver for CPM >> (Continuous-Phase Modulation) signals is defined to be a correlator >> followed by an MLSE detector (Ref. Digital Communications by J. G. >> Proakis). My question, rephrased in the context of this example, is as >> follows: >> >> What difference would it make (in terms of error performance,complexity>> etc etc) if, instead of using the standard approach, I make decision ona>> symbol-by-symbol basis by correlating the received signal during each >> signaling interval with all possible waveforms that could have been >> transmitted, and picking the one that yields the largest correlation? > >That entirely depends on the channel, the signal and the error rate. For>the AWGN, uncoded GMSK BT ~ 0.5 and the error rate ~1e-3, the difference>between MLSE and symbol by symbol detection is ~0.5dB. For the multipath>channels, coded multilevel signals and low BER, the difference can be as>much as 10dB. >I am more interested in understanding the reason behind any performance differential, than mere numbers. Let's say we have an uncoded system in an AWGN channel. The MLSE maximizes the log-likelihood on an entire sequence of symbols and makes a decision on all of them. The symbol-by-symbol detector maximizes the same thing but on each symbol separately. Where is the performance penalty coming from? Is it coming from the (implied) use of Viterbi decoder?>> I hope someone is able to respond to my question this time. My lastone>> didn't elicit a single response :( > >Because you should do your classwork yourself. > >No sir, this is not a homework problem. And it is not a problem I am facing at work either. I am just trying to fill in some of the holes in my understanding of this stuff that I formally tried to learn many many years ago. And I am counting on the kindness of fellow engineers in my pursuit!>Vladimir Vassilevsky >DSP and Mixed Signal Design Consultant >http://www.abvolt.com > > > >
Reply by ●April 7, 20092009-04-07
cyclotron wrote:> Let's say we have an uncoded system in an AWGN channel. The MLSE maximizes > the log-likelihood on an entire sequence of symbols and makes a decision on > all of them. The symbol-by-symbol detector maximizes the same thing but on > each symbol separately. Where is the performance penalty coming from?There is no difference if the channel is memoryless. If the channel has memory, then the signal is no longer the sequence of the independent symbols, but all symbols affecting each other. I.e. it is a multialphabet thing and the optimal decoder is MLSE. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●April 7, 20092009-04-07
> > >cyclotron wrote: > > >> Let's say we have an uncoded system in an AWGN channel. The MLSEmaximizes>> the log-likelihood on an entire sequence of symbols and makes adecision on>> all of them. The symbol-by-symbol detector maximizes the same thing buton>> each symbol separately. Where is the performance penalty coming from? > >There is no difference if the channel is memoryless. If the channel has >memory, then the signal is no longer the sequence of the independent >symbols, but all symbols affecting each other. I.e. it is a >multialphabet thing and the optimal decoder is MLSE. > >Vladimir Vassilevsky >DSP and Mixed Signal Design Consultant >http://www.abvolt.com > > > > >Thanks a lot. That clarified everything about my confusion.
