I've been working with a high-speed (1Ghz+) point-to-point data link
chipset from HP called GLINK (HDMP-1022/HDMP-1024). Both receiver and
transmitter have built-in PLLs used for Transmit Clock generation and
Receive Clock extraction.
The application is PC based so I've been having problems attempting to
use off-the-shelf ATX PC switching power supplies.(i.e. 250W ASTEC) I've
discovered that the supplies have horrible (under and over shoot)
transients (1-100ms@100-500mv) causing loss of lock on the receive end. I
guess I shouldn't expect a lot for 30 bucks!!
The noise seems to correlate with disk accesses mostly (but even mouse
movement shows up), so I'm guessing the supplies have a regulation
problem with variable loads. I haven't gone as far as looking spectrally
at this because I'm not sure what frequencies and DB levels would cause
resonance and/or phase problems in the GLINK . I'm guessing this noise is
coupling in on the TX PLL causing enough/or right frequency jitter so
the receiver PLL loses track. If I cut the boards VCC line and wire in a
linear supply(don't use the PC's VCC, but use the GND), the link stays
solid. The only supply filtering recommendation from HP is the catch-all
0.1uf bypass, but this doesn't seem suitable, so I ask the following.
What is the best way to measure the frequency content of the noise?
Any good reference articles or papers for this situation?
I'm planning a layout change, separating the GLINK's power and ground
layer from the rest of the PCB and use a pi filtering arrangement to
connect the planes, I'm not sure what filter values would be acceptable
or if this is the right approach at all, any ideas?