I kept on getting thesame unstableoutput in LTspice.
One pair of polesseemedto beon the right
half making the system unstable. After you said that you used VCVS, I tried using VCVS and intrestingly found that the output became stable. It can be seenfrom the following image. Fig.1: Left side of the image shows the response with VCVS and the right side shows the LM741 case. It can be seen that in 741 case, at the second pair of conjugate poles the phase seems to lead by 180 degrees rather than reducing. (R2/R1 =100 appeox) Going through the pdf you sent I thought I should reduce that gainof the BPF TF so that, at 0dB the phase is not 180 degrees (which I thought was causing the problem). But if this was the case, it shouldnt have worked for VCVSalso. Stil I went on to try that and found that actually the output became stable. I then decided to try it on MATLAB, I built the transfer function in matlab and tried plotting the the same response. The response was similar to the VCVS case and it remained stable for all gains of the BPF. This made me think that using 741 was causing the problem and I went on to model the 741 transfer function and put it along with the filter transfer function. Simulating this I got theunstable ouput similar to LTspice, which again became stable for lower BPF gains. The plots are shown below. Fig.2 The response with a lesser gain for BPF(R2/R1 =10 approx). Fig.3 : MATLAB output with 741 model (R2/R1 =100 approx) Fig. 4: Matlaboutput with 741 model and lesser gain for BPF(R2/R1 =10 approx). Now, to find how the 741 was creating the problem and reducing the gain makes it stable, I plotted the route locus for both the cases. The Plots are below Fig. 5: Route Locus considering the Buffer to be ideal. Fig.6: Route Locus with 741 as buffer It can be seen that for the case where the buffer is ideal the route locus does not go into Right half of S plane. Usingthe 741 model, which I assumed to have two poles (one at 5 Hz and one at 2MHz) makes the difference; it makes the route locus go into the RHP for higher gains which ultimately explains why I was getting unboundedoutput in LTspice when I used 741. Based on this, I think in the design we shouldalso make sure that the gainof the filter is lesser so that the poles remain in Left half or use a different Op Amp that has higher gain or larger bandwidth, this will give relatively a higher margin for gain; though this also become unstable eventually.