Forums

DVB-H FEC decoder question ...

Started by sudhi November 27, 2005
Hello,

I was looking at the DVB-H standard which has a serially concatenated
encoding with Reed Solomon code (RS(204,188, t=8) from RS(255,239, t=8)
) as the outer code and Convolution code (K=7) as the inner code. At
the receiver, I can think of at least four decoding options,
1. Viterbi followed by RS decoder.
2. MAP (BCJR) followed by RS decoder.
3. MAP followed by soft input RS decoder.
4. Having an iterative decoding with MAP and SISO RS as the component
codes.

I need to answer two questions before I select the decoder.
1. Complexity (DVB-H will be sitting in a handheld device which makes
POWER a big factor.)
2. Performance improvement coming along with the added complexity.

>From the internet I was able to get the answer for MAP decoder. They
say that the MAP is 3-4 times as complex as compared to the standard Viterbi decoder. I am yet to get answers for the following: 1. Can any one tell me by how much factor does the complexity of soft RS decoder and SISO RS decoder increase when compared to the hard input hard output RS decoder. 2. Papers by Koetter-Vardy and Wicker show that soft decoding of RS codes can give upto 6 db improvement for Rayleigh fading channels. Does any one implement soft RS decoders for commercial systems? 3. Knowing that MAP decoder performance is marginally better than Viterbi that too at low Eb/No case, is it worth considering option 2 of decoding. 4. If complexity is not a factor, then it looks to me that Option 4 (iterative decoding) should give the best results. Theoretically, for the above coding scheme, what will be the coding gain. Any other suggestions are most welcome. Thanks, Sudhi
Dang, this has been sitting in the "Drafts" box of Free Agent since I
upgraded it...  :( 

I don't know whether this is still useful but thought I'd send it.


On 27 Nov 2005 06:13:46 -0800, "sudhi" <sudhindra.bellary@gmail.com>
wrote:

>Hello, > >I was looking at the DVB-H standard which has a serially concatenated >encoding with Reed Solomon code (RS(204,188, t=8) from RS(255,239, t=8) >) as the outer code and Convolution code (K=7) as the inner code. At >the receiver, I can think of at least four decoding options, >1. Viterbi followed by RS decoder. >2. MAP (BCJR) followed by RS decoder. >3. MAP followed by soft input RS decoder. >4. Having an iterative decoding with MAP and SISO RS as the component >codes. > >I need to answer two questions before I select the decoder. >1. Complexity (DVB-H will be sitting in a handheld device which makes >POWER a big factor.) >2. Performance improvement coming along with the added complexity. > >>From the internet I was able to get the answer for MAP decoder. They >say that the MAP is 3-4 times as complex as compared to the standard >Viterbi decoder. > >I am yet to get answers for the following: >1. Can any one tell me by how much factor does the complexity of soft >RS decoder and SISO RS decoder increase when compared to the hard input >hard output RS decoder.
Soft RS and SISO RS are both quite a bit more complex than a plain-vanilla RS decoder. How much more complex will depend on implementation details, but I don't know of anyone who has actually built a SISO RS decoder, since they still seem to be primarily in software simulations. Even a soft input RS is difficult to find, and fairly complex. Consider also that to feed a soft-input RS you'll need at minimum a SOVA and a de-interleaver with memory wide enough to hold the soft values for the RS input. Both of those also add complexity in addition to the soft-input RS decoder. I think the reason you don't see it done this way very often is that the gains aren't very large for this additional complexity burden.
>2. Papers by Koetter-Vardy and Wicker show that soft decoding of RS >codes can give upto 6 db improvement for Rayleigh fading channels. Does >any one implement soft RS decoders for commercial systems?
Not to my knowledge, but someone may somewhere. I'm unfamiliar with the paper that you cite, but if it was for an RS-only system in fading then that result may not hold well at all for a concatenated CC-RS system.
>3. Knowing that MAP decoder performance is marginally better than >Viterbi that too at low Eb/No case, is it worth considering option 2 of >decoding.
The depends entirely on what you want to get out of the system, what devices you expect to be using it and where, etc.
>4. If complexity is not a factor, then it looks to me that Option 4 >(iterative decoding) should give the best results. Theoretically, for >the above coding scheme, what will be the coding gain.
How do plan to iterate the decoding? If there is no second encoding like with a Turbo Code I don't know what you'll gain by iterating the decoder. Eric Jacobsen Minister of Algorithms, Intel Corp. My opinions may not be Intel's opinions. http://www.ericjacobsen.org

sudhi wrote:

> Hello, > > I was looking at the DVB-H standard which has a serially concatenated > encoding with Reed Solomon code (RS(204,188, t=8) from RS(255,239, t=8) > ) as the outer code and Convolution code (K=7) as the inner code. At > the receiver, I can think of at least four decoding options, > 1. Viterbi followed by RS decoder. > 2. MAP (BCJR) followed by RS decoder. > 3. MAP followed by soft input RS decoder. > 4. Having an iterative decoding with MAP and SISO RS as the component > codes. > > I need to answer two questions before I select the decoder. > 1. Complexity (DVB-H will be sitting in a handheld device which makes > POWER a big factor.) > 2. Performance improvement coming along with the added complexity. > >>From the internet I was able to get the answer for MAP decoder. They > say that the MAP is 3-4 times as complex as compared to the standard > Viterbi decoder. > > I am yet to get answers for the following: > 1. Can any one tell me by how much factor does the complexity of soft > RS decoder and SISO RS decoder increase when compared to the hard input > hard output RS decoder.
I have not seen any practical realization of SISO RS decoder except for the very simplest RS codes. The SI RS decoders are well known; there are multiple algorithms to do that. The problem with the soft decoding of RS is that the complexity grows exponentially with the size of the alphabet.
> 2. Papers by Koetter-Vardy and Wicker show that soft decoding of RS > codes can give upto 6 db improvement for Rayleigh fading channels.
I have not read this paper however it sounds over optimistic to me. In your case RS is concatenated with convolution code, and I would expect the gain of the SI RS decoder compared to hard RS to be at the order of several tenth of dB.
> Does > any one implement soft RS decoders for commercial systems?
AFAIR soft RS decoding was implemented for Voyager space mission. I haven't heard of any commercial applications.
> 3. Knowing that MAP decoder performance is marginally better than > Viterbi that too at low Eb/No case, is it worth considering option 2 of > decoding.
It depends. Does the gain of ~0.1dB really worth complexity?
> 4. If complexity is not a factor, then it looks to me that Option 4 > (iterative decoding) should give the best results. Theoretically, for > the above coding scheme, what will be the coding gain.
The coding scheme was not optimized for iterative decoding, therefore the expected gain is going to be low. Again, is 0.1 dB worth all of the complications? Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com