ethernetoscillatorphy
I've saw so many reference designs use just a crystal to generate the clock for ethernet phys, though they can be generated with a crystal oscillator too. Such as KSZ9021's 'XI' and 'XO' pin.
Are there some special reason that a crystal is better than a oscillator here? Or just because crystal is cheaper than a oscillator?
Not sure if you are still around or gave up, but in order to create a stable VCXO clock locked to the PAL chroma burst, you need a few timing signals like a "sandcastle" window detector to sample and hold the phase error during this period for the PLL to work.
So use this chip to get your clock and you are good to go.
Since a crystal has an extremely narrow bandwidth (BW) of 1/3000 of the centre frequency, that BW (~1KHz) will determine the best case low pass response delay and lock up time of the Xtal, however once many colour bursts are received or around 10mS of accumulated bursts) , it will stay in sync until the next horizontal line, so that only minor phase corrections can be done during the burst.
It is extremely difficult (but not impossible) to make a PLL lock up in one cycle and then be stable until the next burst. It would require precise measurement of frequency and phase with a 100x clock to get within 1%. using the horizontal sync with 2nS stability. The Horz Sync acts as a freq reference for another PLL loop from the same VCXO with less than 1ppm error and very low phase noise, which sounds easy, but is not cheap.
Found it. MPC is Monitor Products Company, bought out by MMD Components (MMD Monitor/Quartztek Frequency Control Products) in 2002.
970T Datasheet Low-Cost, Full-Size TTL-compatible Clock Oscillator
Best Answer
If the PHY has a crystal driver built in, then using a separate oscillator just for the PHY is rather silly.
A crystal with caps is usually cheaper, smaller, and takes less power than a whole separate crystal oscillator module.