brain death is a terrible thing

It's the golden rule... them's got the gold...

George H.

On the flip side of the coin, I've seen guys blow in to a company with slick stories about "mathematically perfect" systems backed by eggheads with really impressive credentials, whose work couldn't face real-world input data.

"Oh, the textbooks don't cover _that_ case!"

"BCH error correction scheme" ??? What's that? I've heard of holding errors to an "RCH" >:-} ...Jim Thompson

I run into so many "textbook perfect" loons in my work... makes you want to scream and storm out the door ;-) ...Jim Thompson

I once did some performance calculations for a digital facsimile framing sy nchronisation scheme, designed to keep the frames in synch at appreciable n oise levels.

It worked fine in theory, and much worse in the practical tests. When I was asked to explain, it took five minutes to work out that the practical circ uit hadn't implemented the scheme correctly, and rather less to identify th e necessary correction.

Testing real hardware doesn't tell you much about the theoretical scheme bu t it is informative about the real hardware.

Was new to me too. And he's right. It does look complex (complicated)

It has to fail. A channel error that messes up data bits can also mess up BCH bits such as to give an apparently good checksum of bad data. And if the checksum is shorter than the payload, the mapping from payloads to checksums is not unique, so a good message and a garbage message can have the same checksum.

All the math has been done on BCH algorithms; give him the papers and let him figure it out.

A hardware demo is not totally silly. Logic errors happen.

Look it up: en.wikipedia.org/wiki/BCH_code and vocal.com/error-correction/bch-codes/

Surely you are not bitching about this.

BCH is an error correcting code that has the very nice property of being rotation-insensitive. E.g., using a Galois field of m=6, you can handle a code-word of anything from 32-63 bits.

I implemented BCH coding for a client project, as a C++ library, for values up to m=13. We used it with m=6, t=3 (t is the number of single-bit errors that can be corrected), to decode a 33-bit code-word using 18 correction bits. The 51-bit codeword could be shifted up or down within the 63-bit space, giving us 6 bits up and 6-bits down shifting without wrap-around.

The data was a impedance spectrogram with sharp spikes spanning 1.3MHz to 5MHz, which shifted up or down (by a lot!) depending on the temperature of the micro-machined sensor. BCH was a brilliant solution by a friend of mine.

It allowed us to correct the uncertain framing, due to the bit pattern being shifted up or down. The bits at each end are zero (so shifting looks like rotation) so the error-correction still works, and can recover missing framing bits so you can shift the code-word back into the correct position after error-correction.

Clifford Heath.

I guess I must be _really_ old... nobody caught the meaning of RCH

I dated a red head once upon a time... I asked her point blank about RCH... her retort to me was that her mother said I was the type of person she should marry... she said, Blech! ;-) ...Jim Thompson

Had to google for that one.

With the BCH, I forgot to say, we tested it pretty thoroughly against a reference implementation and dataset. Still a possibility of corner cases missed (overflow etc) but it didn't seem likely based on inspection.