Hi Bakshi,
On Wed, 2018-03-28 at 21:37 +0000, Bakshi, Arjun wrote:
1--- The divider(blue) simply tells me when the signal(orange) power has increases more than X times w.r.t. older samples. It isn't exact, but good enough.
No, that's not really the whole picture: it much more tells you about when the older samples were *really* low. y/x grows arbitrarily fast for x->0, but only linearly for increasing y. I've tried to explain why this is a bad idea in my opinion.
2--- However, the divider output is unbounded, so I threshold it (green).
notice that lack of boundedness is definitely a sign of lost information, and you throw away even more by thresholding.
3--- The threshold block's output holds at 1 until the input drops below a value, and doesn't give an impulse(green).
It gives a step. A step contains all frequencies. Especially, a step after your [1,-1] filter gives you an impulse.
4--- So I do the delay and subtract so that only transition points have non-zero values(red).
Exactly!
5--- Of which I want the 1st transition only, which is why I threshold again.
Why threshold? sounds like you want something that has a temporal behaviour, not a value amplitude-dependent behaviour
It isn't exact, but gets the job done.
I think I've clearly laid out my feelings about that!
Best regards,
Marcus
No comments:
Post a Comment