The Negative Feedback Amplifier

A telephone line across a continent needs hundreds of amplifiers in series, and their faults add. Each valve's gain drifts as it ages and each stage adds a little distortion, so a hundred repeaters each drifting one decibel is a hundred decibels of drift. Bell Labs spent the 1920s hand-selecting valves, matching them in pairs and re-matching them every year, and it was not working. Harold Black tried the obvious fix first. His 1923 feedforward amplifier separated out the distortion, amplified it on its own, and subtracted it from the output further down. It works on the bench and it needs two active paths matched to about one per cent and HELD there while the valves age and the room warms up. Nothing in 1923 could do that. On the morning of 2 August 1927, crossing the Hudson on the Lackawanna Ferry, he wrote the other answer on a spare copy of the New York Times. Feed a fraction of the output back to the input in opposition, and the closed-loop gain becomes A over one plus A-beta. Make the loop gain large and that collapses to one over beta - a number set by two resistors, which do not age, do not distort and cost nothing to match. One decision buys four things at once, all divided or multiplied by the same factor: the distortion falls, the gain drift falls, the bandwidth rises, and the amplifier's own characteristics stop mattering. The price is stated just as plainly. To keep forty decibels of gain with forty decibels of feedback you must build eighty and throw half of it away. It also buys a problem nobody had before. A loop can howl, and in 1927 there was no way to predict which one would.