one wire goes to ground, the other goes to pin 11 of the SAA7220
Ignoring for a moment the CD3.5 doesn't use the same chipset as the CD3 (therefore no SAA7220), the clock feed should NEVER, repeat NEVER go to the filter chip first.
It should always go to the DAC, directly to the DAC and not pass go or collect 500ps of jitter on the way
No doubt an improvement can be gained from feeding a better clock elsewhere, but the leads from clock to DAC should be as short as possible, shorter if you can!
In the case of both the CD3 and 3.5 there will be a 74HCU04 unbuffered inverter that formed part of the Pierce oscillator, whether this can be bypassed depends upon whether it is used for further clock distribution, but as a basic guide you need to remove the crystal, and the two load caps connected each end.
See figure 1 in the data sheet link below, RF will be external in the Naim circuit.
http://www.microchip.com/download/appnote/articles/fact001.pdf
In the Naim circuit, IIRC, it's a well-implemented Pierce so it will have a series resistor between the '04 output and the crystal input (this forms the first 90 deg. phase shift for the feedback loop and limits crystal drive level) and probably a large value resistor between input and output.
Interestingly the Pierce oscillator is one of the better types for jitter, but can be improved upon readily with discrete implementation.
Your new clock has three connections, in effect, +Vin, 0V and clock out.
The 0V next to the clock output goes to one of the vacated 0V holes where the caps were removed, the clock output goes to either the input of the '04, or maybe directly to the DAC, should you be prepared to draw the circuit around that area. These wires should be as short as practical.
You then simply need an approximately +12V feed from a convenient point, if it were me I'd seriously consider adding a pre-reg, using an LM317, from one of the internal raw supplies. I use a modified super-reg to provide a v. low noise feed. PSU's are critical for low jitter.
Colin,
The big advantage of the AD744 was the ability to bypass it's very poor output stage to gain improved performance, by taking the output from the compensation pin (pin 5).
The performance of the 744 has been surpassed by a whole ranch of complementary and extra-fast complementary bipolar opamps these days, I'd never recommend a '744 as a good opamp without additional buffering and circuit modification.
Andy.