Stan Pawlukiewicz wrote:> Jim Thomas wrote: > >> Fred Marshall wrote: >> >>> since sgn(x) and sgn(y) are 1-bit numbers ... >> >> >> >> Sorry for clipping out so much context, but according to the OP: >> >> >>>>>> sign(x) = 1 for x>0 >>>>>> 0 for x=0 >>>>>> -1 for x<0. >> >> >> >> By that definition, sign(x) cannot be a one-bit number. >> > > A little hysteresis in the clipper solves that problem.How? there are three states to represent. As Jim Thomas pointed out (most of us were nodding our heads instead of thinking) sgn(x) and sgn(y) are 2-bit numbers (with room to spare). Are you suggesting that the OP should change his procedure? I can't guess if that would be good or not. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
is sign correlation faster than normal correlation on DSPs?
Started by ●October 11, 2005
Reply by ●October 13, 20052005-10-13
Reply by ●October 13, 20052005-10-13
Jerry Avins wrote:> Stan Pawlukiewicz wrote: > >> Jim Thomas wrote: >> >>> Fred Marshall wrote: >>> >>>> since sgn(x) and sgn(y) are 1-bit numbers ... >>> >>> >>> >>> >>> Sorry for clipping out so much context, but according to the OP: >>> >>> >>>>>>> sign(x) = 1 for x>0 >>>>>>> 0 for x=0 >>>>>>> -1 for x<0. >>> >>> >>> >>> >>> By that definition, sign(x) cannot be a one-bit number. >>> >> >> A little hysteresis in the clipper solves that problem. > > > How? there are three states to represent. As Jim Thomas pointed out > (most of us were nodding our heads instead of thinking) sgn(x) and > sgn(y) are 2-bit numbers (with room to spare).In a real clipper, there is always a little noise, even with a dead short across the input. Hysteresis calms the clipper down some. The zero state is something one sees in Matlab, not in the real world. If you want to be mathematically anal about it, the probability that a signal value x=x0, is zero for a Gaussian distribution.> > Are you suggesting that the OP should change his procedure? I can't > guess if that would be good or not.Don't need to guess when you have some actual experince with clipper/correlators.> > Jerry
Reply by ●October 13, 20052005-10-13
Stan Pawlukiewicz wrote:> Jerry Avins wrote:...>>> A little hysteresis in the clipper solves that problem. >> >> How? there are three states to represent. As Jim Thomas pointed out >> (most of us were nodding our heads instead of thinking) sgn(x) and >> sgn(y) are 2-bit numbers (with room to spare). > > > In a real clipper, there is always a little noise, even with a dead > short across the input. Hysteresis calms the clipper down some. The > zero state is something one sees in Matlab, not in the real world. > > If you want to be mathematically anal about it, the probability that a > signal value x=x0, is zero for a Gaussian distribution.Well, the same basic technique that provides hysteresis can provide a dead band between +/- delta. (Like a test for zero in floating point.)>> Are you suggesting that the OP should change his procedure? I can't >> guess if that would be good or not. > > > Don't need to guess when you have some actual experince with > clipper/correlators.I don't, much. Thanks for the education. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●October 13, 20052005-10-13
Jerry Avins wrote:> Stan Pawlukiewicz wrote: >(snip)> > I don't, much. Thanks for the education.This stuff is olde, you can find some of it in the MIT Radiation Lab series.> > Jerry
Reply by ●October 13, 20052005-10-13
Stan Pawlukiewicz wrote:> Jerry Avins wrote: > >> Stan Pawlukiewicz wrote: >> > (snip) > >> >> I don't, much. Thanks for the education. > > > This stuff is olde, you can find some of it in the MIT Radiation Lab > series.I lent my "Principles of Radar" to my son when he was in college, and I haven't seen it since. I thought choke joints and T-R tubes were very clever when I first read about them. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●October 13, 20052005-10-13
Jerry Avins wrote:> Stan Pawlukiewicz wrote: > >> Jerry Avins wrote: >> >>> Stan Pawlukiewicz wrote: >>> >> (snip) >> >>> >>> I don't, much. Thanks for the education. >> >> >> >> This stuff is olde, you can find some of it in the MIT Radiation Lab >> series. > > > I lent my "Principles of Radar" to my son when he was in college, and I > haven't seen it since. I thought choke joints and T-R tubes were very > clever when I first read about them.Tell him to give it back. A Father's job, to teach a son character, never ends. ;)> > Jerry
Reply by ●October 13, 20052005-10-13
"Jim Thomas" <jthomas@bittware.com> wrote in message news:11ksk8nk6330r48@corp.supernews.com...> Fred Marshall wrote: >> since sgn(x) and sgn(y) are 1-bit numbers ... > > Sorry for clipping out so much context, but according to the OP: > > >>>>>sign(x) = 1 for x>0 >>>>> 0 for x=0 >>>>> -1 for x<0. > > By that definition, sign(x) cannot be a one-bit number.Jim, You're correct. I automatically overlooked the zero part because of practical experience. Others have addressed it in terms of hysteresis and noise. So, I stand corrected re: the definition of sgn() and stand on the answer as a good one. So, if sgn() isn't what we want (even though the OP did), then what how, in math, do we *represent* the output of a clipper (suitably scaled and biased or coded to be a single bit 1,0?) sbit()? clip()? I probably should know but don't seem to have an answer on the tip of my tongue. Fred
