I have a 76 MHz (square wave) 12 dbm signal coming out of an Si5351 clock generator that I need to attenuate to -15 dbm, and pass through a 5th-order Chebyshev low-pass LC filter to remove the harmonics (the signal will serve as input to an RF mixer).
Which should be first and which should come next?
As Neil_UK has suggested, an L-pad resistive attenuator, along with coupling capacitor would be best driven by Si5351 output....then followed by a 50-ohm low-pass or band-pass filter.
You might also consider altering drive strength of the final stage Si5351 driver - these are programmable . Doing so will reduce output amplitude a little, but the slower rise/fall times will add to filter attenuation at 2nd and 3rd harmonic.
Spec sheet suggests worst-case rise/fall time is equal at high drive strength (8ma):
1.5ns is faster than needed for a 76 MHz. output!
simulate this circuit – Schematic created using CircuitLab
While I don't see rise/fall time for medium drive strength, rise/fall times will most likely be slower, which can aid the filter's high-frequency attenuation considerably. A lower drive strength will also assist in maintaining a more-constant DC supply voltage. This may be very helpful in preventing internal coupling from one synth chain to another.
When attempting to run Si5351 at reduced drive strength, one should watch out for an increase in even-order harmonics - these are generally not desired when Si5351 is driving a mixer. The cause may be rise/fall times of the internal Si5351 driver's NMOS transistor differing from that of the driver's PMOS transistor.
The mixer may benefit from seeing a good match on its port, so I'd be inclined to put at least some attenuation between the filter and the mixer.
The Si5351 claims to be able to drive a 50 ohm load directly from its outputs, however, be aware that an LC filter, even if designed to be 50 ohms in the passband, is not 50 ohms at all frequencies. Typically, most people implement a lowpass filter with shunt capacitor at each port to ground, as capacitors are cheaper than inductors. This will draw excessive current from a squarewave output.
You could put some of the attenuation before the filter to help with this, or put a diplex filter in front of the LC filter to improve its broadband match, or you could design the LC filter with a series L instead of a shunt C input so it goes high impedance instead of short at high frequency.
a few of the options (there are others)
In the first option, there's no need to make the 5dB attenuator a full '50 ohm on each port' attenuator. It would be better, and kinder to your 5351, to make it an L-pad, driving the filter with 50 ohms, but presenting a higher resistance to your driver.
