Pinout for SED9421
shoppa_classiccmp at trailing-edge.com
Fri Nov 25 08:31:03 CST 2005
Allison <ajp166 at bellatlantic.net> wrote:
> >Subject: Re: Pinout for SED9421
> > From: "Chuck Guzis" <cclist at sydex.com>
> > Date: Thu, 24 Nov 2005 14:19:32 -0800
> > To: cctalk at classiccmp.org
> >Has anyone noticed a significant performance difference between the PLL
> >type data separators (e.g. the WD1691/2143 combo) and the plain-Jane WD9216
> >digital ones? I suspect that the PLL type is more tolerant of drive speed
> >variations, but that hardly matters all that much with modern drives. I
> >don't recall finding a disk that failed to read on a generic 9216-equipped
> >FDC but read fine on the old IBM PC floppy board.
> I've used he 9216 and 9229 and the NEC apnote digital data seperator
> (prom state machine synthetic PLL) as well as a custom NEC ASIC (with
> the prom data sep inside) and they all do very well. Testing against a
> well developed PLL (not the WD animal) showed a small difference after a
> 4 days of continious read and writes (number of fails less than 3).
> You have to do a very large number of reads without external influences to
> see the difference. Power blips and the like were a pain. A small
> difference was seen in the order of one of one failure per 250,000 reads
> vs 1 in maybe 400-500,000 reads. You need a lot more testing than we did
> then to get near 1:10^8 stats with any validity.
> FYI:the average PLL is a bear to build and debug, they required clean
> power and good board layout with ample groundplanes. The digital ones are
> very good, simple to layout and shift rates with only a mux.
Don't neglect: PLL's require analog components of rather tight tolerances
to give consistent behavior from time-to-time and unit-to-unit.
The digital data separator (I heard its designer once refer to it as
a "jerk-locked-loop") has no such tight tolerances and in fact is usually
driven from a crystal oscillator.
The one-shot-with-critical-RC-constant used in early FM data separators
is a good example. With a little tweaking it really works pretty well.
But imagine mass-producing such a beast and training the assembly line
to do the tweaking, as well as field service, as well as ...
(Of course us hackers don't mind!)
More information about the cctech