> I have it tried with a transistor as a driver for every column, when
> tested every columns sink about 40 mA to the GND. BUT whenever
> connected to the PIC16F84 every columns sinks only 3 mA

I think you are spreading the current capabilities of the PIC (as a
primary driver) a little thinly. Even if you do manage to get the
brightness up without using driver transistors, you risk the danger
of exceeding the power rating of the port. You would be better off
with a driver system, with the added advantage that you could
expand the display with few extra demands being placed on the
PIC (apart from timing)

> I wonder if the scanning frequency coused the brightness down.

The scanning frequency itself shouldn't make a difference. What
will affect the brightness is the off to on ratio, and you'll have to
work out what other work the PIC has to do while it is maintaining
regular pulses to the display. For example, shifting new data bits.
Evenly spacing the pulses is also important, so as to keep the
display looking good

> However, when the led connected to frequency generator with
> 128x50 Hz the light strill bright, mean while the scanning frequency
> have adjusted to 128x50 Hz

Which further implies that it's the shortage of PIC power that's
the problem

--
http://www.piclist.com hint: The PICList is archived three different
ways.  See http://www.piclist.com/#archives for details.