FPGA Central

i need to find the step size required to get 1 bit of entropy from a
signal, in order to code it with 1 bit per sample.

this figure should clear up some questions:
http://www1.file-upload.net/30.07.06/etw564.png

so what i did is start from very large step size R and divide it by 2,
calculate the entropy and check if it is under 1.001 until i pass the
1.001 mark. Then i take a very small step size L, multiply it by 2
until the entropy is greater than 0.999 bits. I created two while loops
for each L and R step sizes. The problem is, I get two step sizes which
give an entropie close to one, but not close enough. How would I do it,
so that the entropie gets closer to one, that is, the stepsizes geta
adjusted so that the entropie is closer to 1. I'm perplexed.
I'm using this method of division by 2 because I was told to use it, I
had it explained up to the point where the values pass their bounds
(1.001 and 0.999), after that I was left with a blank.

If somebody has a better method or any suggestions, I would greatly
appreciate it. Thank you,

Paul

[email protected] wrote:
> i need to find the step size required to get 1 bit of entropy from a
> signal, in order to code it with 1 bit per sample.
>
> this figure should clear up some questions:
> http://www1.file-upload.net/30.07.06/etw564.png
>
> so what i did is start from very large step size R and divide it by 2,
> calculate the entropy and check if it is under 1.001 until i pass the
> 1.001 mark. Then i take a very small step size L, multiply it by 2
> until the entropy is greater than 0.999 bits. I created two while loops
> for each L and R step sizes. The problem is, I get two step sizes which
> give an entropie close to one, but not close enough. How would I do it,
> so that the entropie gets closer to one, that is, the stepsizes geta
> adjusted so that the entropie is closer to 1. I'm perplexed.
> I'm using this method of division by 2 because I was told to use it, I
> had it explained up to the point where the values pass their bounds
> (1.001 and 0.999), after that I was left with a blank.
>
> If somebody has a better method or any suggestions, I would greatly
> appreciate it. Thank you,

You might look into the theoretical basis of delta modulation.

Jerry
--
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯ ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

[email protected] wrote:
> i need to find the step size required to get 1 bit of entropy from a
> signal, in order to code it with 1 bit per sample.
>
> this figure should clear up some questions:
> http://www1.file-upload.net/30.07.06/etw564.png
>
> so what i did is start from very large step size R and divide it by 2,
> calculate the entropy and check if it is under 1.001 until i pass the
> 1.001 mark. Then i take a very small step size L, multiply it by 2
> until the entropy is greater than 0.999 bits. I created two while loops
> for each L and R step sizes. The problem is, I get two step sizes which
> give an entropie close to one, but not close enough. How would I do it,
> so that the entropie gets closer to one, that is, the stepsizes geta
> adjusted so that the entropie is closer to 1. I'm perplexed.
> I'm using this method of division by 2 because I was told to use it, I
> had it explained up to the point where the values pass their bounds
> (1.001 and 0.999), after that I was left with a blank.

it looks to me like this is a single threshold step and if that is the
case, there is no step size, just a step location (wherever your DC
reference is for your comparator).

to get the lowest entropy (1 bit per sample), you need to position the
threshold at such a point where there are equal number of samples above
the threshold as are before it. is this what you are doing? it seems
to me that for any PDF, you can find such a point in a method similar
to yours. you're looking for the median.

r b-j

robert bristow-johnson wrote:
> [email protected] wrote:
> > i need to find the step size required to get 1 bit of entropy from a
> > signal, in order to code it with 1 bit per sample.
> >
> > this figure should clear up some questions:
> > http://www1.file-upload.net/30.07.06/etw564.png
> >
> > so what i did is start from very large step size R and divide it by 2,
> > calculate the entropy and check if it is under 1.001 until i pass the
> > 1.001 mark. Then i take a very small step size L, multiply it by 2
> > until the entropy is greater than 0.999 bits. I created two while loops
> > for each L and R step sizes. The problem is, I get two step sizes which
> > give an entropie close to one, but not close enough. How would I do it,
> > so that the entropie gets closer to one, that is, the stepsizes geta
> > adjusted so that the entropie is closer to 1. I'm perplexed.
> > I'm using this method of division by 2 because I was told to use it, I
> > had it explained up to the point where the values pass their bounds
> > (1.001 and 0.999), after that I was left with a blank.
>
> it looks to me like this is a single threshold step and if that is the
> case, there is no step size, just a step location (wherever your DC
> reference is for your comparator).
>
> to get the lowest entropy (1 bit per sample), you need to position the
> threshold at such a point where there are equal number of samples above
> the threshold as are before it. is this what you are doing? it seems
> to me that for any PDF, you can find such a point in a method similar
> to yours. you're looking for the median.
>
> r b-j

I misunderstood the advice given to me. Here is a link to an answer.
Thanks for the replies, greatly appreciated.
http://groups.google.com/group/comp....96c67683201783

Paul