AD5304+PIN work abnormal

Looking at the 'scope picture, the frequency is 50 Hz, a common power line frequency. Its likely that the input is picking up power line interference. To measure low-level input current accurately, careful shielding is need around the input wiring to prevent stray coupling from the power line mains. Be sure to avoid ground loops that include the input circuitry.

Thanks, I use the oscilloscope to make sure my power supply, There is not 50hz sine at flag 20mV of socilloscope. My power supply is lab power, It is work well.

I don't use the ground line in my power supply and socilloscope to avoid the ground loop. There is only one pd on my board to test ADL5304. 

The frequency is 50hz, there is a duty in above picture, The duty is change by PD's current at low range of current. 

what's other can I test for it?

Thanks, I use the oscilloscope to make sure my power supply, There is not 50hz sine at flag 20mV of socilloscope. My power supply is lab power, It is work well.

I don't use the ground line in my power supply and socilloscope to avoid the ground loop. There is only one pd on my board to test ADL5304. 

The frequency is 50hz, there is a duty in above picture, The duty is change by PD's current at low range of current. 

what's other can I test for it?

First, confirm the 50 Hz is line-frequency related. An easy way to do so is set the oscilloscope trigger source to LINE. If the waveform remains stable on screen, it's line-frequency related. 

If line-frequency related, next task is to figure out the way line frequency gets into the circuit. Sometimes its through power supply lines. For this chip, a much more common entry mechanism for line-frequency is through capacitive coupling to the INUM pin, which is very high impedance. You will need electrostatic shielding around this pin, to keep stray line-frequency out. Floating the 'scope and power supply is probably not helping unless you have the circuit board grounded some other way. Also, make sure AC-powered fluorescent or even incandescent light is not shining on the PD, because this will induce 2x line-frequency noise.