FFT timing

I'm pretty sure the early ones were silicon point contact processors.

--
Keith

... snip ...

Don't be silly. You are thinking of selenium rectifiers. :-)

--
 [mail]: Chuck F (cbfalconer at maineline dot net) 
 [page]: 
            Try the download section.


** Posted from http://www.teranews.com **

I wonder what the FFT performance is on this:

John

You don't know anything about electronics and you're too stupid to learn, so why do you post here? You have very little comprehension of the technical material, so reading can't be it.

You don't know about electrolytic rectifiers? Or their optical properties?

John

They also have some strange RF properties.

100K D1 RF ----/\/\------+----------+--->!-- Scope ! ! === C1 ) ! ) L1 ! ) GND ! GND

C1 is a variable capacitor, L1 is a coil that resonates in the AM band when used with C1 and D1 was an electrolytic rectifier.

This is from distand memory:

D1 is constructed by taking a bit of sheet lead and sheet aluminum and spacing them a small distance apart with a few bits of ceramic. A mixture of water and sodiumbicarbonate was poured over the metal and the air bubbles shaken out. This made a device that did rectify somewhat at voltages down near a volt.

The intent was to make an AM radio but what sound I did get from it was really awful and very faint. A scope showed that with a fairly clean 1MHz coming in, if you tuned the capacitor up and down you would find a place where the output jumped about 20mV quite suddenly.

I built one like that too. L1 was hand wound on the core from a roll of TP. The ear piece from a broken hearing aid was the speaker. The Scope showed about 50% efficiency in the rectifier. It worked pretty good for

1955. At least the local station came through

Steven G. Johnson schrieb:

I would be happy with an instruction that converts array-indices into morton-order. Doing the same in software sucks because you end up with a huge chain of dependent instructions.

You can directly do most arithmetic in morton-order, but you end up writing arithmetic routines that are unmaintainable.

With hardware-support we would get rid of this spikey behaviour and improve the overall cache locality of matrix-memory accesses and similar things down to a minimum. Random acceses would still be slow, but anything that as even a hint of locality would benefit a lot.

Technically it has the same cost as any logic instruction because all it does is muxing two or three integers together.

The graphic guys do this for years to improve the efficency of their texture-caches. The database-guys do it since the 70th to remove paging from disk, but for some reason the major CPU-vendors haven't yet found out that their data-caches would benefit from it a lot as well. Or maybe they do know it but don't implement the instruction because C and Fortran can't deal with it.

The TI C64 DSP-family can do that for 2D and 3D array-indexing. Big Kudos for adding the instructions. It improves my code that accesses large two-dimensional arrays by 660% in the average case.

Nils

russ lyttle schrieb:

Ha!

I even remember that the coil need 64 windings for optimal performance.

That circuit was my first radio that I've build with my father in the late 70th. And it worked quite well *)

Nils

*) I must admit we've been living just a couple of miles away from a very strong AM transmitter, so all it took for us to receive radio was to pick up a diode with the left hand and point it into the sky....

All devices received AM including the telephone and the home-stereo. Troubleshooting analog-cirucits was easy as well. Put your finger on the basis of *any* transistor in the circuit. If you can't hear radio the transistor was dead.