Issue is that AD8512 Opamps in Sallen key low pass filters (having G=1) oscillate with nearly 8V swing and high heating of two AD8512 ICs in two boards.
This board works on external 24V, and most experiments that have been done & explained below have been done by isolating the power supply section in the board and feeding external +/-15V with current limiting set on power supplies (in order to avoid damage to the power generation block in board, and the Opamps).
Outputs signals on all four Opamps (on two AD8512 ICs) during heating:
1) With no input given at V+ and V-, signal observations and resultant heating is similar on four legs in two AD8512 ICs.
Signals at TP1 and other corresponding points on similar three legs are: +1.84V and -5.92V (ie 7.76V swing)+2.16V and -6.00V (ie 8.16V swing)+2.08V and -5.84V (ie 7.92V swing)+2.24V and -6.08V (ie 8.32V swing)All signals are approx 1.25 Mhz
Signal at the inverting input of U3 (taken for only one leg out of four) is swing between -3.56V and -440mV (ie 3.12V swing), and freq 970 kHz.
Signal at the non-inverting input of U3 is around -1.8V DC(as expected), and seems like no noise is entering the Opamp. Minus 1.25V * 1.494 (gain) = minus 1.86V
Waveform images taken today are attached and have slightly different swing levels (in fact higher peak to peak), though the frequencies match.
Following experiments were done to help isolate the problem:
2) With V+ and V- shorted, and with this short connected to GND, similar oscillation and AD8512 heating persists.
3) Resistor R6 is lifted, ie AD8221 end of R6 disconnected from the board. In this condition, the sallen key filter is a stand alone circuit between SIG-2 and TP1, with no input or no load at output. With this change, board is powered on.
For this test, 24V given into board power input terminals, and not +/-15V to the board unlike other tests. Current limit on 24V power supply was set to 430mA (current drawn by other boards from 24V power supply used to be 340mA when there was no heating on Opamps).
After power on (with current limit set), 24V from power supply dipped to 2-3V. The higher current draw is (believed as) an indirect confirmation that oscillation and Opamp IC heating persists on the board. There is plan to repeat this test (ie R6 input lifted and grounded) with +/-15V feed to the board.
4) Since the board has a header with jumper option at SIG-1 net, isolated at SIG-1, and fed an external voltage from a new power supply (approx 1.2V) to R4. The oscillation at TP1 which until now was more in the negative region (refer Sl. No. 1 above) moved in the positive direction and got centered nearly around 0V. At this instant, SIG-2 was nearly 100mV with approx 20-30mV ripple. This waveform image was not saved and hence not available. can be repeated, and this seemed to be an useful test.
To add that board is not being kept on for longer duration due to the heating of Opamps even with current limit set for +/-15V power supply.
5) Two 120 ohm resistors soldered from TP1 to GND and from next TP (which interfaces with Opamp on the same AD8512 IC) to GND. Observed just a reduction in the swing levels. Reduced from 2.48V, -6.4V swing to 1.6V, -4.24V swing on TP1. And for the next TP, it reduced from 2.00V, -5.92V swing to 1.52V, -4.4V swing.
The next tests planned are as follows:
6) Isolate the AD8221 end of R6, ground it and power on with current limiting set on external +/-15V supply. Can know for certain if it is oscillation or input noise. If it is due to noise feed, the signal at output would disappear now. If the signals seen earlier were oscillations (as is assumed), the waveform would not disappear.
Reasons to believe that it is not noise feed is that -1.8V is seen at U3 positive pin, and that there is no change in oscillations when V+/V- are connected to GND.
7) Connection changes for capacitor (0.1uF) in the feedback path. Plan is to disconnect C3 pad connected to U3 output (and R9), and to connect this C3 pad to GND.
The circuit would thus have two low pass RC filter stages on input before it reaches U3 positive pin.
8) New 10M or similar higher valued resistor from U3 negative input to GND. This would precisely meet general Sallen key topology. Gain in the filter would thus increase from 1 to 1 + 24.9K/10M. Gain in the existing circuit is 1 + (24.9K/open circuit) ie 1.
9) 24.9K resistor to be shorted or replaced by a 1-2K resistor. Shorting would precisely meet sallen key topology for G=1. Gain becomes 1 + (short/open circuit) = 1.
10) Reduce 31.6K and increase 22nF. This is assuming that very low capacitances in these circuits can lead to errors due to parallel parasitic capacitance of Opamp pins. Not a very convincing argument as Opamp pin capacitances are in pF range, and 22nF doesnt appear low in comparison.
From the tests planned above (6 to 10), 8 to 10 do not look convincing enough to provide leads or a solution, hence would like to know why would this be happening ie Sallen key filter output giving oscillation?
Few documents referred are:1) "Opamps for everyone", from TI; section 16.3.22) MT-222 ("Sallen key filters" by Hank Zumbahlen) from Analog devices
3) A new set of Sallen-Key filter equations by Martin Cano EDN, October 01, 2009 was a good reference, and still looking for more hints in that.
Thank you for the reply.
Active filter wizard was not used in design stage. The specification that we had to meet was a LP filter with fc = 100Hz. It was a redesign being done to solve heating on the earlier LT1352 (Opamps) and LT1167 (IA's). Ten ohm series resistors failing open was also reported in last design.
For both Opamp and IA selection, considered factors like Icc, output current, short circuit current and selected AD8512 and AD8221 respectively. Possibility of output voltage swings which is one reason for higher power dissipation & heating was not considered, or rather not expected, as the circuit is used mostly with DC input.
Answers for the points you asked:
1) cutoff frequency is 100 Hz. The inputs are expected to be DC or slow changing waveforms (10% to 90% ramp in 1 second)
2) There is no load on the AD8512 output other than a TVS diode (SMCJ12CA) to GND between R9 and L1. Oscillation persists even with this TVS diode removed.
3) Using it in product board (PCB mounted)
4) The pole you calculated (1.27Mhz, with stray capacitance on inverting input) should be the second pole I guess, and with first pole around 100Hz.
Planning to carry out experiment #6.