Hello, Can any one point me to paper/book which tabulates the performance of MAP decoder when compared to Viterbi. I see most sites only mentioning that the MAP algorithm outperforms Viterbi at low SNR and high BER condition. Can any one tell me how much is the performance gain by using MAP at low SNR conditions. Thanks Sudhi
MAP Vs Viterbi decoding
Started by ●November 12, 2005
Reply by ●November 12, 20052005-11-12
sudhindra.bellary@gmail.com wrote:> Hello, > > Can any one point me to paper/book which tabulates the performance of > MAP decoder when compared to Viterbi. I see most sites only mentioning > that the MAP algorithm outperforms Viterbi at low SNR and high BER > condition. Can any one tell me how much is the performance gain by > using MAP at low SNR conditions. > > Thanks > SudhiAre you talking about decoding a convolutional code in AWGN channel? John
Reply by ●November 12, 20052005-11-12
sudhindra.bellary@gmail.com wrote:> Hello, > > Can any one point me to paper/book which tabulates the performance of > MAP decoder when compared to Viterbi. I see most sites only mentioning > that the MAP algorithm outperforms Viterbi at low SNR and high BER > condition. Can any one tell me how much is the performance gain by > using MAP at low SNR conditions. > > Thanks > SudhiAre you talking about decoding a convolutional code in AWGN channel? John
Reply by ●November 12, 20052005-11-12
On 12 Nov 2005 01:53:27 -0800, sudhindra.bellary@gmail.com wrote:>Hello, > >Can any one point me to paper/book which tabulates the performance of >MAP decoder when compared to Viterbi. I see most sites only mentioning >that the MAP algorithm outperforms Viterbi at low SNR and high BER >condition. Can any one tell me how much is the performance gain by >using MAP at low SNR conditions.Do you mean to compare Viterbi and MAP decoders for a single convolutional code used by itself? The difference isn't going to be large, and will probably depend on the constraint length. Or do you mean to compare the two when used in an iterative code like a Turbo Code? There the difference is more significant. Eric Jacobsen Minister of Algorithms, Intel Corp. My opinions may not be Intel's opinions. http://www.ericjacobsen.org
Reply by ●November 14, 20052005-11-14
Assuming the convolution coder as specified by the IEEE802.11 or DVB standard (G1=133 and G2 =171) which has a constraint length of 6. The channel is AWGN. I am interested in knowing the performance gain if I replace the Viterbi decoder in the receiver with the MAP decoder which by design is to give me lowest BER. The thought process started when I was looking at the DVB standard. It has a inner convolution code and a outer Reed-Solomon code. If I use a MAP decoder in place of a standard (not SOVA) Viterbi decoder,will the performance of the concatenated system improve. I am not currently thinking of a iterative scheme. I am only going to use the hard outputs from the MAP decoder which i will stream into the RS decoder. So coming back to question what i asked before, what is the performance improvement of MAP over Viterbi when decoding a stream coded using a convolution encoder as specified above assuming the noise is AWGN. thanks Sudhi john wrote:> sudhindra.bellary@gmail.com wrote: > > Hello, > > > > Can any one point me to paper/book which tabulates the performance of > > MAP decoder when compared to Viterbi. I see most sites only mentioning > > that the MAP algorithm outperforms Viterbi at low SNR and high BER > > condition. Can any one tell me how much is the performance gain by > > using MAP at low SNR conditions. > > > > Thanks > > Sudhi > > Are you talking about decoding a convolutional code in AWGN channel? > > John
Reply by ●November 14, 20052005-11-14
I am not looking at a Iterative scheme. My interest is to know the performance difference between Viterbi and MAP (BCJR) when used to decode a sequence which is encoded using a binary convolution encoder (say of constraint length=6).
Reply by ●November 19, 20052005-11-19
Viterbi decoding is Maximum-Likelihood, so it gets lower W( word )ER. But in very low SNR, MAP gets lower BER, because it is optimal for each bit.
Reply by ●January 13, 20062006-01-13
>I am not looking at a Iterative scheme. My interest is to know the >performance difference between Viterbi and MAP (BCJR) when used to >decode a sequence which is encoded using a binary convolution encoder >(say of constraint length=6). > >MAP decoder itself is an iterative scheme. You cannot get the estimate of apriori probabilities of all bits untill you do the first decoding cycle with the assumption of equal distribution.
Reply by ●January 13, 20062006-01-13
On Fri, 13 Jan 2006 02:57:53 -0600, "lanbaba" <lanbaba@gmx.ch> wrote:>>I am not looking at a Iterative scheme. My interest is to know the >>performance difference between Viterbi and MAP (BCJR) when used to >>decode a sequence which is encoded using a binary convolution encoder >>(say of constraint length=6). > >MAP decoder itself is an iterative scheme. You cannot get the estimate of >apriori probabilities of all bits untill you do the first decoding cycle >with the assumption of equal distribution.That's not quite correct, if I understand what you're saying. The MAP or APP decoding algorithm, (two names for the same thing), isn't iterative by itself. You can use a MAP decoder to non-iteratively decode a stream encoded with a convolutional encoder. Performance can be slightly better than that achieved with a Viterbi decoder under certain conditions. MAP decoders are typically used iteratively, though, in things like Turbo Codes where the system iterates between two separate convolutional encodings of the same data (with the order of one being scrambled by an interleaver). The MAP itself is not iterative. Eric Jacobsen Minister of Algorithms, Intel Corp. My opinions may not be Intel's opinions. http://www.ericjacobsen.org
Reply by ●January 17, 20062006-01-17
> You can use a MAP decoder to > non-iteratively decode a stream encoded with a convolutional encoder. > Performance can be slightly better than that achieved with a Viterbi > decoder under certain conditions.Sorry for bringing up an old topic. But, if there is no a-priori information (or for uniform priors), MAP decoding reduces to ML decoding (which is what the Viterbi algorithm does). So unless there is an iterative procedure to estimate the a-priori probabilities, the MAP decoder doesn't do any better than the Viterbi algorithm. So although you CAN perform MAP decoding non-iteratively, it wouldn't be any better than the Viterbi algorithm. - Ravi Srikantiah