Stability Issues with the Hybrid Cascode IF Amplifier  
w7zoi   (2 Dec 07)

It had to happen.   Our IF amplifier (QST, December, 2007) has four stages of gain, and three of them have a reactive element (choke) in the output.   In the remaining case, a transformer is present that has some reactance.  While these inductors are not purposefully tuned, there are stray capacitances in the circuit that tune them somewhere.   Moreover, the inductors themselves have stray C that contributes to a self resonant frequency (SRF) for the part.    With this much gain and this many tuned elements, stability problems should be suspected.    Sure enough, some examination revealed potential problems.

The original circuits we built used toroids in these positions.   These inductors were wound on -43 ferrite material, so were pretty lossy.   After all, this is the material that is often used to calm the EMI problems that we face in our electronic world.   The loss helped to stabilize the amplifier.   We were in the process of preparing a printed board to help folks who want to use the Hybrid Cascode circuit in a receiver application.   In so doing, we had substituted some other RF chokes in the circuit.    The 120 uH inductors that were used in the SMT board were pretty good, even if they were physically large.   They had SRF of 14 MHz, indicating a stray C of only about 1 pF.   Still, I noticed the noise output from a SMT board had quite a bit of peaking down below the nominal 9 MHz amplifier frequency.   Substitution of a lower L value inductor in the differential amplifier output (L4) seemed to calm this problem.

Having seen a problem with a SMT PC board, I wondered if a problem would exist with other variations where molded RF chokes were used.    So the early (no solder mask or silk screen) PC board was fit with the new inductors.    These 47 uH chokes had good enough Q, but had SRF of 8.6 MHz.   This corresponds to a parallel capacitance of 7 pF.   Stray C adds to this to account for a strong noise peak at 5 MHz that I saw in the amplifier output with AGC
turned off.  If I put my finger or a metal tool on a critical point in the circuit, the whole amplifier would go into oscillation at 5 MHz.    This did not happen when AGC was on.

Incidentally, the spectrum analyzer was extremely useful as a tool to study this circuit.   The analyzer was looking at the normal 50 Ohm output from Q8 that would go to a product detector.   The analyzer was spanning from 0 to about 15 or 20 MHz, so showed the 9 MHz signal when one was applied to the IF input.   It would also show the wideband noise drop when AGC is on and input signal is increased.   Peaks in the noise are readily observed.

Fortunately, the problem was fixed with relative ease.   The dominant modification was to add local negative feedback to the differential amplifier.   This stage used a pair of PNP transistors.   Individual emitter bias resistors were used to guarantee nearly equal current in each transistor.   The emitters were then tied to each other with a 0.1 uF capacitor.    The negative feedback was introduced as emitter degeneration and consisted of a small value resistor in series with the 0.1 uF capacitor.    Even 10 Ohms did a lot to help things, and 22 was great.    

The instability was further calmed by decreasing the gain in the IF amplifier.    This was realized by dropping R1 from 2.2 K to 1 K.    The result of these two changes is a noise output that is nearly flat when AGC is off.  The strong peak is no longer there.   The overall gain is still about 58 dB at 9 MHz.   Output with AGC on is still about -35 dBm or less, a reasonable level to drive a diode ring product detector while maintaining very low IMD.

Incidentally, the input network driving Q2 was modified slightly.    The toroid specified in the QST paper was changed from 7.1 uH to 6.8 uH.   This allowed us to use an inexpensive choke in that slot.    The part we picked has a measured Q of about 70 at 9 MHz with SRF way up at 111 MHz, so it performs well in this position.   R7 is dropped to 3.0 K to maintain a good input impedance match with the L-network using the 6.8 uH choke.