Reply by Eric Jacobsen December 8, 20052005-12-08
On Thu, 08 Dec 2005 20:04:20 GMT, Vladimir Vassilevsky
<antispam_bogus@hotmail.com> wrote:

>Eric Jacobsen wrote: > > >>>Does any one know a way to measure the BER after Viterbi or RS decoding at >>>the Rx side if the uncoded data at the Tx side is unknown? >> >> As mentioned already, re-encode and compare (with the uncorrected >> stream) provides a good estimate of the Viterbi decoder's input error >> rate. > >What if the input stream is in the soft decisions?
It usually is, but just taking the sign bit gives an easy comparison for the re-encoder. It's not a problem at all. All you want is the input BER, and the soft information isn't needed (or desired) for that.
>> It's then a pretty simple mapping from input error rate to >> output BER. > >The cost of the best path is the distance between the input stream and >the data decoded. There is no need to re-encode the data since the >information is available already.
That sounds like a nice alternative, but re-encode and compare is quite simple to implement and is a well-known and often used technique. The re-encoder for a convolution code is, as I'm sure you know, quite simple. The results that you can get with the re-encode and compare technique are pretty accurate, usually well within practical user requirements. For people who implement their own Viterbi decoder I'm sure the technique you mention has merit. Often, though, the decoder is purchased either as a discrete part or an IP block, and access to the internal metrics may be difficult or not possible. Eric Jacobsen Minister of Algorithms, Intel Corp. My opinions may not be Intel's opinions. http://www.ericjacobsen.org
Reply by Vladimir Vassilevsky December 8, 20052005-12-08

Eric Jacobsen wrote:


>>Does any one know a way to measure the BER after Viterbi or RS decoding at >>the Rx side if the uncoded data at the Tx side is unknown? > > > As mentioned already, re-encode and compare (with the uncorrected > stream) provides a good estimate of the Viterbi decoder's input error > rate.
What if the input stream is in the soft decisions?
> It's then a pretty simple mapping from input error rate to > output BER.
The cost of the best path is the distance between the input stream and the data decoded. There is no need to re-encode the data since the information is available already. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by Eric Jacobsen December 8, 20052005-12-08
On Tue, 06 Dec 2005 02:35:54 -0600, "lanbaba" <lanbaba@gmx.ch> wrote:

>Hi all, > >Does any one know a way to measure the BER after Viterbi or RS decoding at >the Rx side if the uncoded data at the Tx side is unknown?
As mentioned already, re-encode and compare (with the uncorrected stream) provides a good estimate of the Viterbi decoder's input error rate. It's then a pretty simple mapping from input error rate to output BER. Many (if not most) systems that report their BER on payload data do it this way. Eric Jacobsen Minister of Algorithms, Intel Corp. My opinions may not be Intel's opinions. http://www.ericjacobsen.org
Reply by Vladimir Vassilevsky December 8, 20052005-12-08

john wrote:


>>>You can get a BER estimate based on the average cost of the survived >>>path for Viterbi. >> >>That's a cool idea! So I guess you calibrate the system by injecting >>known noise and measuring the path metric and the BER using a known >>training sequence.
That is brute force approach however yes you can do it this way.
> > One idea is to compare the survivor path to the average of the other > paths. In pure noise, no particular path will dominate so the survivor > path will on average have the same cost as the others. As the signal > rises out of the noise, a lowest cost path should emerge and its cost > compared to the others should give an indication of the BER. Another > idea is to compare the best and worst paths.
You can do it this way also. However you should compare the best path and the next to the best path.
> I don't know how this is > best done in practice but the information should be there to get a BER > estimate.
The cost of the best path is already directly proportional to BER. We are takling about small BERs here. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by john December 8, 20052005-12-08
porterboy76@yahoo.com wrote:
> > You can get a BER estimate based on the average cost of the survived > > path for Viterbi. > > That's a cool idea! So I guess you calibrate the system by injecting > known noise and measuring the path metric and the BER using a known > training sequence. Then the path metric in the case of unkown transmit > data will correspond to a particular BER (on average) using a lookup > table. How does this work with metric normalisation?
One idea is to compare the survivor path to the average of the other paths. In pure noise, no particular path will dominate so the survivor path will on average have the same cost as the others. As the signal rises out of the noise, a lowest cost path should emerge and its cost compared to the others should give an indication of the BER. Another idea is to compare the best and worst paths. I don't know how this is best done in practice but the information should be there to get a BER estimate. John
Reply by December 8, 20052005-12-08
> You can get a BER estimate based on the average cost of the survived > path for Viterbi.
That's a cool idea! So I guess you calibrate the system by injecting known noise and measuring the path metric and the BER using a known training sequence. Then the path metric in the case of unkown transmit data will correspond to a particular BER (on average) using a lookup table. How does this work with metric normalisation?
Reply by Vladimir Vassilevsky December 6, 20052005-12-06
lanbaba wrote:

> Does any one know a way to measure the BER after Viterbi or RS decoding at > the Rx side if the uncoded data at the Tx side is unknown?
You can get a BER estimate based on the average cost of the survived path for Viterbi. For the syndrome decoding of RS, the average number of corrected errors per block gives you the BER estimate. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by Nic December 6, 20052005-12-06
Hi Lanbaba,

I think, this is exactly what the author in this paper also does, he
just calls it Pseudo-channel BER (pcBER), so not much new info for you,
I am sorry :(.

Anyhow, much luck with your search..

Best wishes

Nic

Reply by lanbaba December 6, 20052005-12-06
Hi Nic,

thank you for the reference. To estimate the BER before Viterbi I just
re-encode the Viterbi output and compare the encoded bits with the Viterbi
input. Sometimes this estimate is called Viterbi error flag rate (VEFR).

Cheers, Lanbaba
Reply by Nic December 6, 20052005-12-06
Hi,

I have once read a paper on how to estimate the BER without reference:
http://www.ebu.ch/en/technical/trev/trev_274-schramm.pdf

This is probably not exactly what you are looking for, as it estimates
the BER before viterbi decoding, but you might be able to work out a
solution from there.

Best wishes

Nic