SSB Crystal filter response

Michael VE1LEB has been scratch-building a uBITx.   This is the first time he has built a crystal filter.  The photo below shows the response curve of the 12MHz SSB filter output by his PHSNA:

To select the crystals, Michael used the K8IQY crystal test fixture and chose the closest ones to 12MHz from a bag of 50 inexpensive computer crystals.

He is not sure whether his PHSNA is accurately calibrated or how much error there is in the frequency readings. However, the -20dBm passband is less than 2.5khz.   He received a new bag of 100 crystals, and was wondering if he should take the time to  

Rod KS6SM commented,

“In the ubitx, the crystal filter is driven/terminated in a 200 ohm impedance. It is likely your PHSNA is 50 ohms.  The passband ripple you are seeing can be severely influenced by driving/terminating impedances.

“Are you measuring the filter with the transformers at each end, or are you going direct to the crystal connections?

“It is normal for the passband to be below 12 MHz. On my ubitx, the BFO is at 11,997,117 Hz, so that will give you a sense of how far below 12 MHz the passband is.”

It turned out that Michael was measuring the filter ‘naked’, without the transformers.  He measured it again but through the transformers, and the result is  much “softer”:

Satish VU2SNK said, “In both the curves without and with transformer your filters loss appears to be around -7 db.   In my opinion this is bit too high -3db is just right up to -5db is acceptable according to the experts, what type of capacitors are used in this filter? The multi layer ceramic capacitors
are really of low loss. If you use the modern SMD ones in my opinion that will reduce the loss in the filter.”

Allison KB1GMX also suggested, “One warning when measuring narrow filter use a very slow sweep. At narrow resolutions if the sweep is too fast the curve will not match the actual. It is as much detector response time as its filter delays. When too fast the it will appear tilted, and when you slow down you will see more accurate result.

The filter insertion loss in the 4 to 7 db range sounds correct. Though I might be better if higher Q capacitors are used we are not talking a 3DB difference. Also dId you calibrate out the transformers first? That can add a DB of loss sometimes more.”

In responding to the suggestions from Allison and Satish,  Michael said, “I’m using chinesium SMD 1206 capacitors.

“I realize now that I had not accounted for loss in my cables, and perhaps I’m not appreciating fully the implications of my test equipment. I’m a dopey artist, not an engineer! When shorted, the cables out-of and into the PHSNA present a 3.5 dBm loss. Each transformer adds 1.9 dBm loss, so double that and add the loss in the cables will represent a total loss of 7.3 dBm. When I bias the PHSNA for the 7.3 dBm loss and attach the filter (including the transformers on each end), I get this result, which puts me well inside the 4-7 dBm loss that you mention.”