Talk:Turbo code
From Wikipedia, the free encyclopedia
Contents |
[edit] Latency
"For satellite use, this is not of great concern, since the transmission distance itself introduces latency due to the limited speed of light." - Can the writer of this sentence justify this? Considering how fast the speed of light is, I don't think this can really contribute to latency. - Yongqli
- Well, looking at the geosynchronous article here on Wikipedia gives the distance to a GEO satellite (approx. 36000 km), and the speed of light is 300000 km/s. The round-trip distance is twice the altitude, or 72000 km, and 72000 / 300000 = 0.24 s (or 240 msec), a significant delay in many types of communication. europrobe 10:38, 14 January 2006 (UTC)
[edit] The Shannon Limit
User:HughSW -- hats off to your addition How Turbo codes work -- really nice work! technopilgrim
[edit] Compared with Reed-Solomon
I am interested to know if turbo code is more efficient than Reed-Solomon error correction for those areas that Reed-Solomon is particularly used for. For example, given the same number of additional bits, is turbo code better able to handle errored signals? Is turbo code better able to handle missing signals? Is turbo code well suited to 'bursty' errors? Also, on modern desktop CPUs, which is most time-efficient for encoding and decoding? --Yamla 18:08, 2005 Mar 21 (UTC)
In terms of encoding efficiency, Turbo Codes are the best known (as mentioned in the first paragraph). Bursty errors are usually handled by interleaving/rearranging the bits (as in read solomon's usage on CD's). Sorry I can't tell you which is most time-efficient. 194.106.59.2 20:37, 28 Mar 2005 (UTC)
[edit] How turbo codes work
I think the emphasis of the How turbo codes work section is wrong: the focus is on soft-bits while this is not what Turbo codes made different, it is the fact that two codes in parallel (interleaved) are used. The soft decoding was already known before ([1]: "It is well known that soft decoding is better than hard decoding") Emvee 20:43, 16 July 2005 (UTC)
- This is true. (LDPC improves upon turbo coding for the same reason.) --Piet Delport 20:34, 6 January 2006 (UTC)
-
- Agreed; that was completely wrong. Soft-decision decoders are used routinely with just about every kind of code there is. I've removed that section. 66.30.10.35 01:09, 29 January 2007 (UTC)
[edit] Nitty gritty
Wow this article has the lame term "nitty-gritty" twice. (Once with a hyphen and once without, heh.)
- Yeah, that seriously has to go.[update] Alright, I got rid of it.--Soban 15:00, 1 August 2006 (UTC)
[edit] Typical number of iterations?
The Wikipedia article says "typically in 15 to 18 cycles", but both articles referenced at the end say "typically 4 to 10". Which is more correct?
My gut feel is that the number depends on some external factor, like the number of errors, in which case the statement should probably be replaced entirely with an explanation of what factors influence the iteration count, and why. --Piet Delport 20:15, 6 January 2006 (UTC)
Sometime the definition of what constitutes an 'iteration' is blurred. Technically an interation is the combination of two component decode operations (MAP/Log-MAP/SOVA etc), one using the non-interleaved parity bits, and one using the interleaved ones (i.e. 6 iterations involves 12 component decodes). Sometimes people use the term iteration to just mean number of component decodes. It is not uncommon for a turbo decoder to feature an 'early termination' option, where you compare the hard output of the previous N iterations (where N is generally between 2 to 3). If the outputs are the same (i.e. they have converged), you can generally stop decoding with very little effect on BER. In high SNR channels this is good as the decoder generally quickly converges to the right answer (i.e. more iterations don't tell you anything new). Typically its also good in low SNR channels for the opposite reason (the decoder quickly converges to the wrong answer so more iteration doesn't help either). In mid-SNR channels the decoder usually run to its maximum number of iteration anyway. The maximum number of iterations required for a given application is generally derived from the expected SNR of the channel and the a required bit or frame error rate target for the link. --miterdale 12:51, 15 January 2006 (UTC)
