Anomalous behavior of RF power amplifier made using Cree

[arwen16]

Hello,

We designed a class AB PA using CG2H40010F (VGS=-2.4V, ID=250mA) on Rogers 5880 (Er=2.2). We did the S-parameters measurement and they were in good agreement with the simulation. We did notice post measurement that our board was slightly warmer (we did use a heat sink of 40x60mm dimension). We waited for it to cool down before moving to the power measurements. We used a SG, a preamp, a 60dB attenuator and an SA to measure the power output. On connecting the PA in the setup, there was a sudden spike in drain current (~1A). We quickly turned it off and disconnected the preamp from the i/p and the attenuator at the o/p. We checked the DC and it was stable. When we tried reconnecting to the preamp, there was a spike in current again. We then removed the preamp and decided to just connect the input of the PA to the SG with RF off. We encountered the same issue. We did notice there is a small gap between the board and the heat sink even after applying thermal paste.

We are fairly new to designing PA and are still understanding the practical implementation of one. Kindly help.

 

[volker@muehlhaus]

Your bias supply looks suspicious to me, I don't see suffient measures to decouple the solder pad from the RF line.

Newbies often simulate with a voltage source (=RF short) at that node, but the real world supply line are very different from an RF short. Did you check how isolation from the RF line (transistor input) to the DC pads? I wonder if your spike is oscillation due to RF feedback across the red DC wires which seem to touch each other.

 

[arwen16]

Hello,

Could you tell us how to check for isolation? The wires are not touching each other; it's just in the picture.

 

[volker@muehlhaus]

Could you tell us how to check for isolation?

I would use simulation and place one port at each end of the bias path, then check S21 of that path. We want high isolation, something like -30dB for S21. And of course we want that path to be high impedance at the end where it touches the RF line.

 

[arwen16]

We checked and the S21 was less than 10dB. We tried to check for oscillations by limiting the current and saw no spike in the SA.

 

[volker@muehlhaus]

We checked and the S21 was less than 10dB.

Not sure if you mean isolation better than 10dB or worse than 10dB?

In any case, that is not much. You can try adding a ferrite bead to your wires, next to the pad, just to see if the spike (oscillation?) disappears.

 

[arwen16]

We checked and the S21 was less than -10dB from 1-3GHz and below -30dB in the rest. We tried to check for oscillations by limiting the current and saw no spike in the SA. We then gave a 25dBm input signal to check if we get any power at our desired frequency (1.8 GHz, 2.4 GHz and 3.5 GHz) and we did get some output power though the drain voltage dropped significantly; unable DC

 

[D.A.(Tony)Stewart]

How many nH/mm and fF/mm did you assume for parasitics, crosstalk and mutual coupling?

If you don't know then you might want to rethink your layout and examine commercial LNAs and see how to control SNR, impedances and specify L, C for coupling biasing.

BTW This example is already obsolete.

--- Updated May 28, 2024 ---

When you choose microstrip, without ground planes on top, you have a difference in dielectric constants, Dk on top air-side and underside with a different wave velocity as well as exposure to parasitic crosstalk and instabilities above some threshold. That is a bad idea for mm waves.

What made you ignore the design features provided by Macom?

 

[arwen16]

T

How many nH/mm and fF/mm did you assume for parasitics, crosstalk and mutual coupling?

If you don't know then you might want to rethink your layout and examine commercial LNAs and see how to control SNR, impedances and specify L, C for coupling biasing.

View attachment 191127

BTW This example is already obsolete.

--- Updated May 28, 2024 ---

When you choose microstrip, without ground planes on top, you have a difference in dielectric constants, Dk on top air-side and underside with a different wave velocity as well as exposure to parasitic crosstalk and instabilities above some threshold. That is a bad idea for mm waves.

What made you ignore the design features provided by Macom?
View attachment 191128

Thank you so much for your input. We are new at this and since we have no clear guidelines, we are making mistakes. We will definitely look into the ground plane. Is this the reason our DC is not stable?

--- Updated May 28, 2024 ---

I would really appreciate if anyone can share a good and reliable source material for practical PA design.

 

[vfone]

In the demonstration schematic of CG2H40010F from the datasheet, there are two decoupling electrolytic capacitors, C8=10uF and C17=33uF.

I don't see any of them in your circuit.

 

[arwen16]

In the demonstration schematic of CG2H40010F from the datasheet, there are two decoupling electrolytic capacitors, C8=10uF and C17=33uF.

I don't see any of them in your circuit.

Yes, we did not use them. I do wish to understand though. We have seen many papers that implement bias networks of their own, i.e. not following the demonstration schematic.

Is it then absolutely necessary to follow the schematic?

 

[vfone]

Put in this way. First, follow the schematic and all the recommendations of the manufacturer. Later, after everything is working fine and performance meet the datasheet, you can try removing components which YOU consider should not be where they are, and then check if the circuit is still in datasheet parameters.

 

[arwen16]

Put in this way. First, follow the schematic and all the recommendations of the manufacturer. Later, after everything is working fine and performance meet the datasheet, you can try removing components which YOU consider should not be where they are, and then check if the circuit is still in datasheet parameters.

Thank you so much. I will do better in my next design. Is there no way to save this one though? Or test it somehow? We could not see any oscillations in the SA.

 

[volker@muehlhaus]

Is there no way to save this one though?

You didn't show us what you did, except for a photo where we can't identify components or values.

 

[arwen16]

You didn't show us what you did, except for a photo where we can't identify components or values.

We have used a 50 ohm resistor in the gate, a parallel combination of 50 ohm and 3.5 pf for stability at the input. We have used 100nF and 3.5pF as RF bypass capacitors.

--- Updated May 30, 2024 ---

Could you please tell me what these circles imply in the demonstration circuit?