We am having trouble getting full gain and the maximum RF output power our HMC3573. I am operating the device at 8447 MHz, and am biasing the device with Vdd = 8.1 VDC and a drain bias current of 1200 mA. I am only able to get 27.1 dB of gain and +33.6 dB of RF power out of the device and I would like to get over 3 W out of the device, as is shown in the datasheet. We have also purchased an evaluation board for the device and can not get full power out of that either.
I have verified that the gates and drain pins are adequately decoupled and that the drain voltages do not droop when RF power is applied. The device is mounted on a 30 mil RO4350C core with solid ground plane on the bottom that is then clamped to an aluminum block with thermal grease between them. There appears to be adequate heat sinking occurring since the block heats up well and no loss of power is seen as the device and block heats.
While I am not seeing full power out of the device, when I increase the RF input power and record the output power and drain currents, I do get similar curves as shown in the Power Compression @ 8 GHz plot on page 5 of the datasheet.
Any ideas why we are unable to get full gain and power out of the device, especially the one mounted on the evaluation board that we purchased? What else should I be checking?
In order to better undestand what you are seeing can you send me these two:
1. Gain and Pout vs Pin at 8GHz?
2. s2p file
I do not have any S-parameters measured with a VNA since its a high-power amp, so measuring S22 is not straight forward. I can measure |S11| and |S21| with my bench set up that uses directional couplers and RF power sensors. As such, here are some Gain, Pout, and Drain Current vs. Pin numbers from testing done at 8.447 GHz. Results were very similar at 8.0 GHz. I don't have the reflected power numbers on me at the moment.
Pin (dBm) Pout (dBm) Gain (dB) Idd (mA)
0 27.10 27.10 1195 (oddly, I see the drain current drop slightly from the DC bias of 1200 mA when RF is applied)
1 28.19 27.19 1196
2 29.00 27.00 1205
3 29.78 26.78 1227
4 30.51 26.51 1257
5 31.19 26.19 1293
6 31.84 25.84 1334
7 32.52 25.52 1371
8 33.17 25.17 1395
9 33.48 24.48 1330
10 33.62 23.62 1313
From these numbers, I see the trends indicated in the datasheet, but the overall gain and Pout are just low. Let me know if there is any other data you would like to see.
- Can you share data for eval board?
- Assuming that this part does not oscillate, you can measure with NA because you use small level signal at input
lets say -25dBm.
But even forget output RL accuracy if you insert attenuators at output but you can measure S11 and see if there is some big mismatch at input.
I do not know for board RO4350C I only know for RO4350B, please clarify?
If so, are you sure that your lines are close to 50ohms?
- I see you used up to input power of 10 dBm so I am curious about your setup, is this power level directly from SG,
if so, harmonic content might not be good for these levels depending on SG used.
Hi Ivan, thanks for the feedback.
I think I am good to to. I doubled up on the decoupling capacitors next to the Vdd pins and that seemed to help - may have been current starving the device a little. I am now able to get a saturated power of +34.9 dBm out of my board.
Also, I did measure the S11 with the VNA and saw about 12 dB of return loss, so the lines may not be a perfect 50 Ohms and/or there may be some mismatch at the input and output interfaces from the PCB to the SMA connectors.
For the 50 Ohm transmission lines, I am using a grounded co-planar waveguide (GCPW) structure that is constructed using a 6.6 mil thick Rogers RO4350B core with a center trace width of 14 mils and spacing to the top side ground plane of 18 mil. The copper is 0.5 oz thick (~0.7 mil) that has been ENIG plated, so total metal thickness is probably ~1.5 mils thick on the top layer. For the SMA connectors, in order to try to minimize any mismatch in transitioning from GCPW to coaxial cable, I am using a CPW connector, part number 142-0761-861, made by Cinch Connectivity Solutions.
I will continue to try some things, but there does not seem to be any major design flaws. It looks like it could just be some combination of minor tolerance things adding up.