I had a crack at following various Qucs and other tutorials relating to LNA design. The amplifier is intended for 1.7GHz HRPT signal reception. I hope someone might be able to make some comments about how to bring it to unconditionally stable as K is below 1 for much of my region of interest in frequency. K is >1 at high frequency.
I picked what I think should be a suitable transistor, an BFP842ESD and went through ngspice simulation to find a suitable operating point for low NF.
Here are the noise, stability circles and reflection coefficients of the transistor at the chosen operating point and @1.7GHz
Initially I chose to (mis)match the input for the reflection coefficient given by the intersection of the NFmin circle with the highest GA circle value (toward the exterior of the SG input stability circle) and subsequently recomputing with the input match in place and then conjugate matching the output seemed to go OK but with a double inductor L match for the output - I presume the calculator tool decided a capacitor was too small to be realisable.
Here is the schematic and Smith Chart plots for stability circles, reflection coefficient and noise figure circles and a frequency sweep of K, NF, S21, S11, post application of both input and output matching networks:
I noted that S11 was outside the input stability circle and S22 inside but very close to the output stability circle. Output seems well matched to 50 ohm.
Being prudent, I took the second intercept point of NFmin and GA at a lower gain and repeated the matching exercise yielding:
At the expense of some gain I go with option 2 (in fact, I have tried both but post is rather long already) since the L network contains capacitors that can act as DC blocks it's convenient. I use my previous ngspice results to add a suitable bias network, using the inductors as the DC path and RF decoupling them so RF sees the original matching network. See schematic:
and the plots:
I note that S11 is outside the input stability circle. S22 is inside the output stability circle @f_0 . The frequency sweep clearly shows K < 1 around my area of operation. Referencing Infineon's app note AN_1805_PL32_1806_113119, I have derived a circuit fairly close to what they have in structure.
I wonder about the value of my C4. I chose it to appear short to RF. In the app note they have a few pF (different matching network values too but not such order of magnitude different) I notice that moving it to a value where L2, C4 are series resonant @ around 3.5GHz (guessing that the series resonant frequency of this pair to ground is significant somehow...) I see improvement in K. What is going on here?
Also, in the app note they have put shunt 160R direct across the output port - I see also improves the K situation at the cost of some gain.
Here is the frequency sweep with C4=10pF and R10 activated:
Hmm ran out of attachments. Anyway, K < 1 only between 1 and 2 GHz at the cost of a fair amount of gain, now 17.8dB @1.7GHz and slightly worse NF = 0.515 dB.
I'll try degenerating the emitter with some inductance and see what that brings. Anything else I can try? I know it's not unconditionally stable but is it likely to be OK?
Also, regarding realisation, would I be better off using double stub matching with microstrip at these frequencies since the C's and L's are exact values and design is very sensitive to these and I could in theory cut PCB trace to tune? Or better start the matching network from scratch with known nominal value L's and work from there?