'USB Clock Issues (48MHz +\- 0.25%) [OT]'
Looking at a USB device. First looked at the Phillips PDIUSBD11 USB
interface with I2C but had lots of troubles sourcing it. I am now looking at
the Texas Instruments TUSB2140B 4port hub with embedded fuinction accessible
via I2C. This product appears very similar to the Philips PDIUSBH11, but
with slightly less intelligence (it doesn't handle device classes for you
for instance). I chose this as I can actually source it (Digikey at US$5.24
quantity 1, 500+ in stock) and it should be possible to migrate from this to
one of the Philips products without much difficulty as they are both I2C
(this only does 400kbps I2C, Philips does 1Mbps). I may want to migrate as
the Philips model if I don't want the hub (Texas Instruments have no hubless
version) or I want fatser access. Also, the TI device has only 2 endpoint
pairs as it's embedded endpoint is low speed only (the hub ports are of
course full speed).
My problem is with clocking. The USB specs state a clock accuracy in a slave
device of 0.25%. The Philips specs don't state a required clock tolerance,
the Microchip specs for the upcoming USB PICs refer you to the USB specs and
the TI specs state the 0.25% tolerance. To achieve this tolerance, TI
suggest sourcing resonators from Murata specially sorted for the board
capacitance of your PCB as standard Murata resonators are 0.5% and the best
they offer generally is 0.3%. This is a problem, firstly Murata aren't big
here in Australia, secondly I would like to test a circuit without having to
ship it to Murata and thirdly I doubt custom sorting for your PCB comes
cheap. The TI device requires a 48MHz clock which makes it even harder as
few manufacturers have 48MHz crystals\resonators (Murata do).
Can anyone suggest an appropriate clock source (it can take clocks as well
as resonators\crystals), perhaps a PLL and a crystal as most other USB ICs
use (i.e. MChip, philips)? I would assume that a 0.3% resonator from Murata
should do the trick at least for testing. If the specs state 0.25% it'll
prob. work fine with up to 0.4% and if the resonators state 0.3% they should
mostly be 0.25%, but the 0.3% varieties are harder to source (anyone know a
distributor who ships to Australia?) and either way I wouldn't like to use
them for anything more than testing and even then. Could someone who's
completed a USB design let me know what clock they used and whether there
were any problems. The TI specs say clock jitter can prevent enumeration (it
actually says numeration), does this mean after enumeration clock jitter is
not so imprtant or does it mean you can't do anything but of course
enumeration is the first step.
> Looking at a USB device. First looked at the Phillips PDIUSBD11 USB
> interface with I2C but had lots of troubles sourcing it.
They still don't learn. They have to issue small quantities/samples to
evaluate and design a product before we can order the 1000+ which was
the MOQ when I first looked at the PDIUSBD11! They are getting better
however, I was forced to purchase 27 units. (Adjusted, more reasonable
MOQ when Adilam took over.) Considering your in Australia, if you make
it worth my while, (Cost ($4.58 inc tax) + P&P), I can send you over
some for you. Email me if you are interested.
Why can't you use a Crystal? Is it because you can't source one at
The USB specs state under 7.1.9, that the Data Signaling Rate tolerance
must be +/- 0.25% (2500ppm) for full speed and 1.5% (15000ppm). I'm not
sure how the internals of the hub works being Full Speed, but
considering your embedded endpoints are low speed, I would assume the
resonators would be o.k. < 1.5%.
Crystals on the other hand are more accurate. They typically range
around +/-20 to 50 PPM with a temperature stability of +/- ~50 PPM. Even
with worse case scenario of crystal capacitance loading etc., you would
still make the spec with room to move.
I've been using a generic 12Mhz Crystal on my PDIUSBD11 which uses a PLL
x4 to get the 48Mhz Clock. Early stages were on breadboard (Philip's
Data Sheets aren't all that good. You have to play) thus even with all
the stray capacitance there were no noticeable problems.
> Can anyone suggest an appropriate clock source (it can take clocks as well
> as resonators\crystals), perhaps a PLL and a crystal as most other USB ICs
> use (i.e. MChip, philips)?
If you are having trouble finding a 48Mhz Crystal, you could use a
Crystal Oscillator Module e.g. Farnell 483-394. I was going to try
something like this with NatSemi's USBN9602.
Interfacing the PC -
Information on Serial / RS232 Ports, Parallel, ECP, EPP
Ports, IRQ's, and AT Keyboards, USB & Device Drivers.
More... (looser matching)
- Last day of these posts
- In 1999
, 2000 only
- New search...