Imler.dyndns.org

Moore’s Cow

Person A: So, newegg is offering a 2 gig Creative Zen Stone Plus for $30.

Person B: Man, I paid $40 for 1 gig last year.

Person C: Me too.

Person D: Those prices are going down, but the price of milk is skyrocketing.

Person B: Sadly, Moore’s Law does not apply to cattle.

2 comments May 16th, 2008

CaramellDancen

So, some of the guys at improv were unaware that I have a blog. They were momentarily worried that I might be posting embarrasing things about them.

As if.

This blog exists so that I can blither about the latest stupid meme to completely hijack my frontal lobes. In this case, Caramell Dancen Be warned: they’ve set anime characters to do a stupid dance to fast paced swedish bubblegum pop.

I can’t get this crap out of my head.

1 comment May 6th, 2008

Stupid Reed-Solomon Tricks

I just stumbled across a pretty neat thing you can do with a Reed-Solomon encoded bitstream. If the data you’re feeding into the RS system has sidebar information about erasures, and you can model the proportion of erasures to undetectible bad symbols (as a function of raw bit error rate), you can use the observed number of erasures in your input stream to reliably guess if the data in your input buffer is gibberish.*

It turns out that not running known gibberish through your ECC significantly drops your false positive rate.

Yeah, I’m building a packet protocol on top of a lower speed channel with a bit reliability of only 80% - 85%. The guys at the bar tell me that my beer-fueled rants on the topic are, “Impressive”.

*To illustrate: A RS(10,2) system can correctly decode if 8 > 2*errors + erasures. Let’s say that you know an erasure is twice as likely as an error if the Bit Error Rate is constant over your sample area. (not a bad assumption for small sample areas) Given 4 erasures, you can expect about 2 errors, and 8 > 2*2 + 4 won’t ever correct, and running RS will probably give you garbage.

Add comment May 1st, 2008