Forums

New algorithms in MLSE?

Started by Verictor October 26, 2013
Hi,

Anyone is aware of state search algorithms in MLSE (maximum likelihood sequence equalization) other than the Viterbi?

Thanks
On Sat, 26 Oct 2013 14:18:09 -0700 (PDT), Verictor
<stehuang@gmail.com> wrote:

>Hi, > >Anyone is aware of state search algorithms in MLSE (maximum likelihood sequence equalization) other than the Viterbi? > >Thanks
Sequential decoder? Can I ask why you think you need MLSE? Eric Jacobsen Anchor Hill Communications http://www.anchorhill.com
On Sunday, October 27, 2013 1:02:07 AM UTC-7, Eric Jacobsen wrote:
> On Sat, 26 Oct 2013 14:18:09 -0700 (PDT), Verictor > > <stehuang@gmail.com> wrote: > > > > >Hi, > > > > > >Anyone is aware of state search algorithms in MLSE (maximum likelihood sequence equalization) other than the Viterbi? > > > > > >Thanks > > > > Sequential decoder? > > > > Can I ask why you think you need MLSE? > > > > > > Eric Jacobsen > > Anchor Hill Communications > > http://www.anchorhill.com
For some reason, my Group didn't get updated to see new posts. Sequential decoder is used on convolutional codes, as far as I know. Can it also be used on non-coded communication? In particular, combined equalization and maximum likelihood state estimate? I was referring to channel estimate along with incoming sequence detection. There may be ways to incorporate CIR estimate in the sequential decoder (or someone had done that already but I don't know). I was asking to see if someone knows new techniques to minimize the search. Why need MLSE? I think a sequence base, not symbol-by-symbol, approach is optimal as intuitively, we have more degree of information. Thanks
On Tue, 29 Oct 2013 10:31:44 -0700 (PDT), Verictor
<stehuang@gmail.com> wrote:

>On Sunday, October 27, 2013 1:02:07 AM UTC-7, Eric Jacobsen wrote: >> On Sat, 26 Oct 2013 14:18:09 -0700 (PDT), Verictor >>=20 >> <stehuang@gmail.com> wrote: >>=20 >>=20 >>=20 >> >Hi, >>=20 >> > >>=20 >> >Anyone is aware of state search algorithms in MLSE (maximum likelihood s= >equence equalization) other than the Viterbi? >>=20 >> > >>=20 >> >Thanks >>=20 >>=20 >>=20 >> Sequential decoder? >>=20 >>=20 >>=20 >> Can I ask why you think you need MLSE? >>=20 >>=20 >>=20 >>=20 >>=20 >> Eric Jacobsen >>=20 >> Anchor Hill Communications >>=20 >> http://www.anchorhill.com > >For some reason, my Group didn't get updated to see new posts.=20 > >Sequential decoder is used on convolutional codes, as far as I know. Can it= > also be used on non-coded communication? In particular, combined equalizat= >ion and maximum likelihood state estimate?
It's possible, but sequential decoding is used primarily when a long constraint length can be tolerated in favor of better performance. Modulation-based trellis patterns tend to be short, which favors the Viterbi algorithm.
> I was referring to channel estim= >ate along with incoming sequence detection.
Soft-input sequential decoders exist.
> There may be ways to incorporat= >e CIR estimate in the sequential decoder (or someone had done that already = >but I don't know). I was asking to see if someone knows new techniques to m= >inimize the search.=20 > >Why need MLSE? I think a sequence base, not symbol-by-symbol, approach is o= >ptimal as intuitively, we have more degree of information.=20 > >Thanks
Eric Jacobsen Anchor Hill Communications http://www.anchorhill.com
On Tuesday, October 29, 2013 2:33:13 PM UTC-4, Eric Jacobsen wrote:
> On Tue, 29 Oct 2013 10:31:44 -0700 (PDT), Verictor > > <stehuang@gmail.com> wrote: > > > > >On Sunday, October 27, 2013 1:02:07 AM UTC-7, Eric Jacobsen wrote: > > >> On Sat, 26 Oct 2013 14:18:09 -0700 (PDT), Verictor > > >>=20 > > >> <stehuang@gmail.com> wrote: > > >>=20 > > >>=20 > > >>=20 > > >> >Hi, > > >>=20 > > >> > > > >>=20 > > >> >Anyone is aware of state search algorithms in MLSE (maximum likelihood s= > > >equence equalization) other than the Viterbi? > > >>=20 > > >> > > > >>=20 > > >> >Thanks > > >>=20 > > >>=20 > > >>=20 > > >> Sequential decoder? > > >>=20 > > >>=20 > > >>=20 > > >> Can I ask why you think you need MLSE? > > >>=20 > > >>=20 > > >>=20 > > >>=20 > > >>=20 > > >> Eric Jacobsen > > >>=20 > > >> Anchor Hill Communications > > >>=20 > > >> http://www.anchorhill.com > > > > > >For some reason, my Group didn't get updated to see new posts.=20 > > > > > >Sequential decoder is used on convolutional codes, as far as I know. Can it= > > > also be used on non-coded communication? In particular, combined equalizat= > > >ion and maximum likelihood state estimate? > > > > It's possible, but sequential decoding is used primarily when a long > > constraint length can be tolerated in favor of better performance. > > Modulation-based trellis patterns tend to be short, which favors the > > Viterbi algorithm. > > > > > I was referring to channel estim= > > >ate along with incoming sequence detection. > > > > Soft-input sequential decoders exist. >
I've never actually implemented a SISO sequential decoders. Dosn't soft sequential decoder become more complex so that it loss its advantage of simpler implementation complexity? I think some texts say sequential decoder is usually hard decoder. There are second opinions on this?
> > > > There may be ways to incorporat= > > >e CIR estimate in the sequential decoder (or someone had done that already = > > >but I don't know). I was asking to see if someone knows new techniques to m= > > >inimize the search.=20 > > > > > >Why need MLSE? I think a sequence base, not symbol-by-symbol, approach is o= > > >ptimal as intuitively, we have more degree of information.=20 > > > > > >Thanks > > > > > > > > Eric Jacobsen > > Anchor Hill Communications > > http://www.anchorhill.com
On Tue, 29 Oct 2013 21:47:54 -0700 (PDT), Verictor
<stehuang@gmail.com> wrote:

>On Tuesday, October 29, 2013 2:33:13 PM UTC-4, Eric Jacobsen wrote: >> On Tue, 29 Oct 2013 10:31:44 -0700 (PDT), Verictor >>=20 >> <stehuang@gmail.com> wrote: >>=20 >>=20 >>=20 >> >On Sunday, October 27, 2013 1:02:07 AM UTC-7, Eric Jacobsen wrote: >>=20 >> >> On Sat, 26 Oct 2013 14:18:09 -0700 (PDT), Verictor >>=20 >> >>=3D20 >>=20 >> >> <stehuang@gmail.com> wrote: >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >> >Hi, >>=20 >> >>=3D20 >>=20 >> >> > >>=20 >> >>=3D20 >>=20 >> >> >Anyone is aware of state search algorithms in MLSE (maximum likelihoo= >d s=3D >>=20 >> >equence equalization) other than the Viterbi? >>=20 >> >>=3D20 >>=20 >> >> > >>=20 >> >>=3D20 >>=20 >> >> >Thanks >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >> Sequential decoder? >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >> Can I ask why you think you need MLSE? >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >>=3D20 >>=20 >> >> Eric Jacobsen >>=20 >> >>=3D20 >>=20 >> >> Anchor Hill Communications >>=20 >> >>=3D20 >>=20 >> >> http://www.anchorhill.com >>=20 >> > >>=20 >> >For some reason, my Group didn't get updated to see new posts.=3D20 >>=20 >> > >>=20 >> >Sequential decoder is used on convolutional codes, as far as I know. Can= > it=3D >>=20 >> > also be used on non-coded communication? In particular, combined equali= >zat=3D >>=20 >> >ion and maximum likelihood state estimate? >>=20 >>=20 >>=20 >> It's possible, but sequential decoding is used primarily when a long >>=20 >> constraint length can be tolerated in favor of better performance. >>=20 >> Modulation-based trellis patterns tend to be short, which favors the >>=20 >> Viterbi algorithm. >>=20 >>=20 >>=20 >> > I was referring to channel estim=3D >>=20 >> >ate along with incoming sequence detection. >>=20 >>=20 >>=20 >> Soft-input sequential decoders exist. >>=20 > >I've never actually implemented a SISO sequential decoders. Dosn't soft seq= >uential decoder become more complex so that it loss its advantage of simple= >r implementation complexity? I think some texts say sequential decoder is u= >sually hard decoder. There are second opinions on this?
Soft-input sequential decoders have been around a long time and are generally of lower complexity than (or comparable complexity to, depending on constraint lengths) a typical soft-input Viterbi decoder. But I was mostly thinking of detection or decoding rather than an MLSE Equalizer (brain fart on my part, as you did say equalizer). A MAP decoder (aka APP decoder) is not an MLSE, but it can be used for equalization in the same manner as a Viterbi algorithm, or joint channel estimation and equalization. I doubt a sequential decoder would be desirable for channel estimation or equalization unless the channel delay spread covered a large number of symbols, i.e., the equalizer was hugely complex. So, probably not practical.
>>=20 >>=20 >> > There may be ways to incorporat=3D >>=20 >> >e CIR estimate in the sequential decoder (or someone had done that alrea= >dy =3D >>=20 >> >but I don't know). I was asking to see if someone knows new techniques t= >o m=3D >>=20 >> >inimize the search.=3D20 >>=20 >> > >>=20 >> >Why need MLSE? I think a sequence base, not symbol-by-symbol, approach i= >s o=3D >>=20 >> >ptimal as intuitively, we have more degree of information.=3D20 >>=20 >> > >>=20 >> >Thanks >>=20 >>=20 >>=20 >>=20 >>=20 >>=20 >>=20 >> Eric Jacobsen >>=20 >> Anchor Hill Communications >>=20 >> http://www.anchorhill.com >
Eric Jacobsen Anchor Hill Communications http://www.anchorhill.com
On Tuesday, October 29, 2013 10:21:33 PM UTC-7, Eric Jacobsen wrote:
> On Tue, 29 Oct 2013 21:47:54 -0700 (PDT), Verictor > > <stehuang@gmail.com> wrote: > > > > >On Tuesday, October 29, 2013 2:33:13 PM UTC-4, Eric Jacobsen wrote: > > >> On Tue, 29 Oct 2013 10:31:44 -0700 (PDT), Verictor > > >>=20 > > >> <stehuang@gmail.com> wrote: > > >>=20 > > >>=20 > > >>=20 > > >> >On Sunday, October 27, 2013 1:02:07 AM UTC-7, Eric Jacobsen wrote: > > >>=20 > > >> >> On Sat, 26 Oct 2013 14:18:09 -0700 (PDT), Verictor > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> <stehuang@gmail.com> wrote: > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> >Hi, > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> > > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> >Anyone is aware of state search algorithms in MLSE (maximum likelihoo= > > >d s=3D > > >>=20 > > >> >equence equalization) other than the Viterbi? > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> > > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> >Thanks > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> Sequential decoder? > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> Can I ask why you think you need MLSE? > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> Eric Jacobsen > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> Anchor Hill Communications > > >>=20 > > >> >>=3D20 > > >>=20 > > >> >> http://www.anchorhill.com > > >>=20 > > >> > > > >>=20 > > >> >For some reason, my Group didn't get updated to see new posts.=3D20 > > >>=20 > > >> > > > >>=20 > > >> >Sequential decoder is used on convolutional codes, as far as I know. Can= > > > it=3D > > >>=20 > > >> > also be used on non-coded communication? In particular, combined equali= > > >zat=3D > > >>=20 > > >> >ion and maximum likelihood state estimate? > > >>=20 > > >>=20 > > >>=20 > > >> It's possible, but sequential decoding is used primarily when a long > > >>=20 > > >> constraint length can be tolerated in favor of better performance. > > >>=20 > > >> Modulation-based trellis patterns tend to be short, which favors the > > >>=20 > > >> Viterbi algorithm. > > >>=20 > > >>=20 > > >>=20 > > >> > I was referring to channel estim=3D > > >>=20 > > >> >ate along with incoming sequence detection. > > >>=20 > > >>=20 > > >>=20 > > >> Soft-input sequential decoders exist. > > >>=20 > > > > > >I've never actually implemented a SISO sequential decoders. Dosn't soft seq= > > >uential decoder become more complex so that it loss its advantage of simple= > > >r implementation complexity? I think some texts say sequential decoder is u= > > >sually hard decoder. There are second opinions on this? > > > > Soft-input sequential decoders have been around a long time and are > > generally of lower complexity than (or comparable complexity to, > > depending on constraint lengths) a typical soft-input Viterbi decoder. > > > > > > But I was mostly thinking of detection or decoding rather than an MLSE > > Equalizer (brain fart on my part, as you did say equalizer). A MAP > > decoder (aka APP decoder) is not an MLSE, but it can be used for > > equalization in the same manner as a Viterbi algorithm, or joint > > channel estimation and equalization. > > > > I doubt a sequential decoder would be desirable for channel estimation > > or equalization unless the channel delay spread covered a large number > > of symbols, i.e., the equalizer was hugely complex. So, probably not > > practical. >
That is a candid suggestion. When I asked, I was thinking other ways to do VA. Thanks anyway.
> > >>=20 > > >>=20 > > >> > There may be ways to incorporat=3D > > >>=20 > > >> >e CIR estimate in the sequential decoder (or someone had done that alrea= > > >dy =3D > > >>=20 > > >> >but I don't know). I was asking to see if someone knows new techniques t= > > >o m=3D > > >>=20 > > >> >inimize the search.=3D20 > > >>=20 > > >> > > > >>=20 > > >> >Why need MLSE? I think a sequence base, not symbol-by-symbol, approach i= > > >s o=3D > > >>=20 > > >> >ptimal as intuitively, we have more degree of information.=3D20 > > >>=20 > > >> > > > >>=20 > > >> >Thanks > > >>=20 > > >>=20 > > >>=20 > > >>=20 > > >>=20 > > >>=20 > > >>=20 > > >> Eric Jacobsen > > >>=20 > > >> Anchor Hill Communications > > >>=20 > > >> http://www.anchorhill.com > > > > > > > Eric Jacobsen > > Anchor Hill Communications > > http://www.anchorhill.com