What is the effect of my finger? Attenuators abnormal !

Hi Everyone! I have designed a downconverter for 5.05GHz~5.75GHz, of which the input power range is -60dBm~+30dBm. I have to put some attenuators to set the fit mixer level. In other words, I designed two separated ways to accomplish the level control before the mixer and the circuit is shown as below (You can zoom on it to see the details!).
The sma port in the bottom is the input port. U44 is the first attenuator, which is HMC624, a 0.5dB-step 31.5dB total attenuation attenuator. The U46 is a SPDT switch, which is in response with the other SPDT U49. U45 is the amplifier for weak signal. C136 way is used for strong signal. U52 is the second attenuator of HMC624.
Now it comes the trouble when I test the circuit. Each attenuator is in normal state when tested individually (simply like the first is 0dB, 10dB, 20dB, 30dB set and the second is set 0dB and so forth). But when the two attenuators are cascaded in attenuation to get more than 30dB attenuation, the circuit can not respond rightly. The total attenuation can not be controlled at all. For example, when set 45 dB in control software, the final state is about 35dB (I test it in the IF output of my circuit, which is in response with the circuit). What is worse, there is no law to control it rightly. This is not funny, because there maybe some electromagnetic radiation to overwhelm the energy transferred through the microstrip. However, when I put some absorptive materials in upside of the circuit, when my fingers touch the materials, the normal attenuation can be reached (To tell the truth, the attenuation is different with different place I touch a bit forcebly); When my fingers leave the materials, the normal attenuation get gone! When I closed up with the shielding upside-board, the normal attenuation get still gone!
So what is the problem with my circuit? How to solve it? What is the effect of my fingers? Can I make some equivalent circuit to compensate it? Best regards! Any reply is welcome.

Where is your design layout and S11, S22 test results? It appears stray leakage is affecting results and finger dielectric shunts low levels to ground depending on standing wave position.

Here is the datasheet of the attenuator and the circuit pic.

- - - Updated - - -

Here is the layout and spec. Sorry late for internet speed!

- - - Updated - - -

It may be the stray leakage. If there is standin wave, the attenuation should be equal to a pai (PI) pad. So the S-parameters should be better for bigger attenuation state. All the frequency band exists this phenomenon. So I don't think this is the S-parameter problem.

- - - Updated - - -

Sorry I dont test the S11 and S22 parameters for it is part of a whole circuit (I mean S22).

I don't see controlled 50R impedances with shielding for the ATT in out paths. Perhaps I'm reading your layout incorrect.

SunnySkyguy said:

I don't see controlled 50R impedances with shielding for the ATT in out paths. Perhaps I'm reading your layout incorrect.

Click to expand...

The effect you observed often happens if attenuators of >20 dB are cascaded.
One way is the leakage over the PCB or the flat-mounted devices, other leakage can be expected by control lines and DC power line.
All those ways must be carefully blocked so that no leakage is < 60 dB, over the full frequency range of attenuating stages.
As you indicated, inserting a lossy foam did some effect, so make sure no wave can propagate through the cavity. Check grounding and use a lossy foam anywhere except "live" 50-Ohm lines.

I wonder if saulbit is still fighting with the problems reported more than 3 years ago? https://www.edaboard.com/threads/202117/

Regarding the layout screenshot in "What is Wrong with the circuits", I think it would be necessary to look at the details thorougly, that's difficult with just a photo. I'm not sure what the other experts think, but I would need the full gerber data and stack description. jiripolivka's absorbing foam suggestion should be tried first.

YES FvM, it appears SOULBIT did not appreciate that Antenna effects and controlled impedance requirements start at about 1% of the wavelength or even less, such as this when the "Noise is 60dB greater" than the signal in close proximity.

Worse yet, the dielectric properties in FR4 changes about 1GHz and up, with reduction in dielectric constant and loss tangent rising. So that at 5GHz, FR4 is unusable. There is no indication what was used.

There is no evidence of stripline or microstrip on this board , hence wrong impedance tracks between attenuators will give poor return loss and transmission loss when cascaded and also be more susceptible to radiation.

Yes this is the way I have made some progress in 4GHz successfully. I just wonder why the microwave magnetic field radiate so much more for 6GHz than the same way in 4GHz. This circuit is not layouted by me. In some case I would like to make some metal shielding between the attenuators. Now I have to solve the problem. I simulated the S2P files in ADS, and the open circuit (such as open microstrip stub in HFSS), both results of which support the design. But in fact it fails.
3 years ago, the reason for attenuation abnormal is soldering, the ground pad of the HMC624 should be soldered as large as possible.
The pcb material is RO4350, not FR4. The absorbing foam is tried, getting a little improvment for about 2 dBs.
I wonder why the energy comes out, which should be absorbed by the PI pad in the attenuator ICs. However, the S11 parameters can not be 60dB, which is almost ideal.

- - - Updated - - -

The impedance is calculated in the ADS linecalc. The RO4350 is about 1.07mm, with a 50 Ohm width about 1.07mm. The board is 4 layers, top layer for smts and microstrips, second layer is gnd plane, third layer for power and control lines and bottom layer for gnd plane. The via is buried for bottom layer to be as a whole gnd.

- - - Updated - - -

SunnySkyguy said:

YES FvM, it appears SOULBIT did not appreciate that Antenna effects and controlled impedance requirements start at about 1% of the wavelength or even less, such as this when the "Noise is 60dB greater" than the signal in close proximity.

Worse yet, the dielectric properties in FR4 changes about 1GHz and up, with reduction in dielectric constant and loss tangent rising. So that at 5GHz, FR4 is unusable. There is no indication what was used.

There is no evidence of stripline or microstrip on this board , hence wrong impedance tracks between attenuators will give poor return loss and transmission loss when cascaded and also be more susceptible to radiation.

Click to expand...

I know the antenna effects and I think it is the cause of the trouble. But why 4.5GHz works normally at the same time 6GHz works abnormally, of which the frequency difference is only about 1.5GHz. I simulated open microstrip stubs (two stubs, sepereated by the gnd pad to maximize the effect of radiation, there might some antenna radiates more than open stubs?) ,the result is worse about 15dB from 4.5GHz to 6GHz.
What is wrong with my simulation? some wrong concepts? Please point out for me. When I study in colleage, the main problem is never cover octave bands. This design does not fit the octave band. I would offer the HFSS model soon.

This is the HFSS simulation model of open stubs.

Consider adding an extra copper layer or putting lower level signals on other side of a middle ground layer or using a tinned brass shield over top.this was essential for 1GHz radios we made.