In both types of amplifiers there is, of course, always the chance that the output circuit of one tube may be coupled to and induce some effect in the input circuit of one of the earlier tubes of the series. This will be amplified and result in a greater induction. In other words, in a circuit where there is large amplification, there is always the difficulty of avoiding a feed-back of energy from one tube to another so that the entire group acts like an oscillating circuit, that is "regeneratively." Much of this difficulty can be avoided after experience. If a multi-stage amplifier is to be built for a current which does not have too high a frequency the "capacity effects" and the other difficulties due to high-frequency need not be seriously troublesome. If the frequency is not too high, but is still well above the audible limit, the noises due to variations in battery currents need not bother for they are of quite low frequency. Currents from 20,000 to 60,000 cycles a second are, therefore, the most satisfactory to amplify. Suppose, however, one wishes to amplify the signals from a radio-broadcasting station. The wave-length is 360 meters and the frequency is about 834,000 cycles a second. The system of intermediate-frequency amplification solves the difficulty and we shall see how it does so. [Illustration: Fig 128] At the receiving station a local oscillator is used. This generates a frequency which is about 30,000 cycles less than that of the incoming signal. Both currents are impressed on the grid of a detector. The result is, in the output of the detector, a current which has a frequency of 30,000 cycles a second. The intensity of this detected current depends upon the intensity of the incoming signal. The "beat note" current of 30,000 cycles varies, therefore, in accordance with the voice which is modulating at the distant sending station. The speech significance is now hidden in a current of a frequency intermediate between radio and audio. This current may be amplified many times and then supplied to the grid of a detector which obtains from it a current of audio-frequency which has a speech significance. In Fig. 128 I have indicated the several operations. We can now see why this method permits sharper tuning. The whole idea of tuning, of course, is to arrange that the incoming signal shall cause the largest possible current and at the same time to provide that any signals at other wave-lengths shall cause only negligible currents. What we want a receiving set to do is to distinguish between two signals which differ slightly in wave-length and to respond to only one of them.