First, if you want to measure the filter characteristic, you should cut the connection to the amplifier IC2A, otherwise its output impedance will affect your measurement.

Second, to reach 4 GHz, I'd suggest contacting the board by cutting a coax cable. Solder the outer braid to the circut ground as near as possible to the signal trace you want to contact. Solder the coax center conductor to the signal trace. Cut off any signal trace extending beyond where you soldered in the coax.

For example:

(image source)

(source)

Depending on the details of your board design, you might be able to hack in an SMA connector, but the direct-connected coax will work quite well up to 5 or 6 GHz.

The sig-gen has 50 Ohm output, and the built-in scalar analyzer has 50 Ohm input. Do I need to use a buffer circuit terminated in 50 Ohms (and perhaps with a 2x gain) before and after the DUT

No, the buffer will prevent you seeing any reflections from the filter, which are an important part of characterizing the filter.

The SNA should have some means of calibrating out cable reflections. It won't be as good as a VNA because the SNA doesn't measure as much information as the VNA does, but it will be better than blocking yourself from getting reflection data from the filter.

Note

Your filter is apparently designed to be driven by NE5532, an op-amp. This means it's meant to be driven with a low-impedance source rather than a 50-ohm source. You will need to do some post-processing to account for this after measuring its response with your 50-ohm network analyzer.

First, if you want to measure the filter characteristic, you should cut the connection to the amplifier IC2A, otherwise its output impedance will affect your measurement.

Second, to reach 4 GHz, I'd suggest contacting the board by cutting a coax cable. Solder the outer braid to the circut ground as near as possible to the signal trace you want to contact. Solder the coax center conductor to the signal trace. Cut off any signal trace extending beyond where you soldered in the coax.

For example:

(image source)

(source)

Depending on the details of your board design, you might be able to hack in an SMA connector. Or, if you think ahead when doing the board design you can add a footprint for an SMA connector that is normally not loaded. But when it comes down to it, the direct-connected coax will work quite well up to 5 or 6 GHz.

The sig-gen has 50 Ohm output, and the built-in scalar analyzer has 50 Ohm input. Do I need to use a buffer circuit terminated in 50 Ohms (and perhaps with a 2x gain) before and after the DUT

No, the buffer will prevent you seeing any reflections from the filter, which are an important part of characterizing the filter.

The SNA should have some means of calibrating out cable reflections. It won't be as good as a VNA because the SNA doesn't measure as much information as the VNA does, but it will be better than blocking yourself from getting reflection data from the filter.

Note

Your filter is apparently designed to be driven by NE5532, an op-amp. This means it's meant to be driven with a low-impedance source rather than a 50-ohm source. You will need to do some post-processing to account for this after measuring its response with your 50-ohm network analyzer.