Intersymbol interference, not noise, imposed a ~2400 bits-per-second 'distortion barrier' on early-1960s voiceband telephone data

In his 1965 Bell System Technical Journal paper, Robert W. Lucky located the ceiling on digital transmission over voice-grade telephone lines in the channel's distortion, not its noise floor. He wrote that "present data rates on voice telephone channels are limited to about 2400 bits per second... the nonuniform transmission characteristics of the channel cause what might be termed a distortion barrier, prohibiting faster transmission" (Lucky 1965, BSTJ 44(4), p.548).

The mechanism named here is intersymbol interference: a voiceband channel's non-flat amplitude and non-linear phase response smears each transmitted pulse in time so that it bleeds into the slots of its neighbours. Push the symbol rate up and the smearing — not thermal or additive noise — is what corrupts the decision at the receiver. The barrier is therefore a linear-distortion limit, in principle correctable by inverting the channel's response, which is precisely the opening Lucky's adaptive equalizer exploited (claim-lucky-1965-adaptive-equalizer-steepest-descent-transversal-filter).

The framing matters for how the telecom line of adaptive-filter work parallels the neural-net line: Widrow's group left neural nets for adaptive signal processing on channels like this one (claim-widrow-abandoned-multilayer-training-until-1985-backprop), where the object to be learned is an inverse filter for a physical medium rather than a classifier's weights.