I am currently working on a polyphase filter, which is implemented using opamp-RC. The op-amp is written using veriloga with a DC gain of 50 dB and unity gain frequency of 500 MHz. The schematic is shown in Fig. 1.

A cmdmprobe is added in the feedback path of the in-phase opamp and the value of cmdm is set to -1. The obtained loop response is shown in Fig. 2. We can see that when the loop gain phase is 0 at 14.04 MHz, the loop gain is 20.99 dB. It implies an oscillation condition.

The in-phase opamp has a differential feedback loop gain response shown in Fig. 3, when the cross-coupling resistors between I and Q are removed. This loop gain response is easy to understand.

To verify the stability, I added a voltage impulse in one feedback path, shown in Fig. 4. The voltage impulse has both rising and falling slope as 10mV/5ns. The resulted transient response is shown in Fig. 5. We can see that the loop is actually stable.

This design has actually been fabricated and measured. Indeed, there is no oscillation. My question is that why the stability analysis shows that oscillation is possible at 14.04 MHz (zero loop phase and greater than unity loop gain)? Any comment/discussion is appreciated.

Hi Frank, thanks for your quick comment.

Frank Wiedmann wrote on Dec 12^th, 2012, 1:09am:


diffstbprobe was introduced in MMSIM10.1. Currently I only have access to MMSIM06.21.431_LNX86, where there is only cmdmprobe.

Frank Wiedmann wrote on Dec 12^th, 2012, 1:09am:


The amplifier in in-phase path without I/Q cross-coupling has no stability problem, as shown in Fig. 3. Thanks.

Frank Wiedmann wrote on Dec 12^th, 2012, 1:09am:


Please see Fig. B1 and Fig. B2 for the "global feedback" loop response, where the cmdmprobe was added in the coupling path from I to Q. We can see that the global feedback is stable, as suggested by the PM of 30 deg and GM of 35 dB. So my doubt is that when the global feedback is stable, shall we care about the "local feedback" stability as shown in Fig. 2? Are we supposed to have "absolute" stability in every feedback loop? The simulation of the "local feedback" loop response should be correct. I derived the local feedback loop response and obtained the bode plot using matlab, which is close to Cadence stability results.

Is the "local feedback instability" a special phenomenon for polyphase filter and can be ignored?

Thank you.

CMChan wrote on Dec 12^th, 2012, 12:37am:


Zero loop gain phase and greater than unity loop gain magnitude does not mean instability. That would be true where the loop gain has only poles and no zeros.

nrk1 wrote on Dec 12^th, 2012, 7:52am:

Assuming the op-amp differential feedback resistor as Rlp and differential feedback capacitor as Clp. There are two zeros at low frequencies: one at 1/(2*pi*Rlp*Clp) and another is close to the first one. Due to negative feedback, the loop gain phase increases from -180 deg. Yes, the point at 14.04 MHz in Fig. 2 is mostly contributed by the two zeros. Thanks.