On the inadequacies of loop gain in broad-band, multiloop feedback amplifiers

Experience in communications engineering has repeatedly shown that loop gain can be an unreliable basis for the assessment of stability margins in feedback amplifiers. There are both theoretical and practical reasons for this. A simple identity provides a sounder theoretical basis, and provides a novel way of looking at old problems. The presence of any unstable modes of the closed-loop amplifier can be predicted with more assurance from open-loop measurements by examining a product of functions derived from the identity rather than a single-loop gain. Further improvements in stability assessment are possible by using the data base of the embedding network method [1]. The loop gains computed from the data base can be compared with those derived directly from two-port s-parameter measurements at the corresponding nodes, thus avoiding the use of circuit diagrams. Results on several practical amplifiers are quoted, which show that plots of loop gains can fail to predict closed-loop instability. In another actual case, a loop gain failed to predict conditional stability. The correct predictions in each case were obtained using the method described in [1].