Suppose we set up a tuned circuit formed by a coil and a condenser and try it out for various frequencies of signals. You know how it will respond from our discussion in connection with the tuning curve of Fig. 51 of Letter 13. We might find from a number of such tests that the best we can expect any tuned circuit to do is to discriminate between signals which differ about ten percent in frequency, that is, to receive well the desired signal and to fail practically entirely to receive a signal of a frequency either ten percent higher or the same amount lower. For example, if the signal is at 30,000 cycles a tuned circuit might be expected to discriminate against an interfering signal of 33,000. If the signal is at 300,000 cycles a tuned circuit might discriminate against an interfering signal of 330,000 cycles, but an interference at 303,000 cycles would be very troublesome indeed. It couldn't be "tuned out" at all. Now suppose that the desired signal is at 300,000 cycles and that there is interference at 303,000 cycles. We provide a local oscillator of 270,000 cycles a second, receive by this "super-heterodyne" method which I have just described, and so obtain an intermediate frequency. In the output of the first detector we have then a current of 300,000--270,000 or 30,000 cycles due to the desired signal and also a current of 303,000--270,000 or 33,000 cycles due to the interference. Both these currents we can supply to another tuned circuit which is tuned for 30,000 cycles a second. It can receive the desired signal but it can discriminate against the interference because now the latter is ten percent "off the tune" of the signal. You see the question is not one of how far apart two signals are in number of cycles per second. The question always is: How large in percent is the difference between the two frequencies? The matter of separating two effects of different frequencies is a question of the "interval" between the frequencies. To find the interval between two frequencies we divide one by the other. You can see that if the quotient is larger than 1.1 or smaller than 0.9 the frequencies differ by ten percent or more. The higher the frequency the larger the number of cycles which is represented by a given size of interval. While I am writing of frequency intervals I want to tell you one thing more of importance. You remember that in human speech there may enter, and be necessary, any frequency between about 200 and 2000 cycles a second. That we might call the range of the necessary notes in the voice. Whenever we want a good reproduction of the voice we must reproduce all the frequencies in this range.