494 BELL SYSTEM TECHNICAL JOURNAL 



measuring devices were in good condition as far as could be reasonably ex- 

 pected over the range of speeds. The distortion measured without channel 

 filters ranged from 2.5 per cent at 60 w.p.m. (23 d.p.s.) up to 10 per cent at 

 170 w.p.m. (65 d.p.s,). 



Swing 



Some measurements were made on the complete frequency-shift arrange- 

 ment over the same range of speeds, using several values of abrupt frequency 

 swing from ±15 cycles to ± 55 cycles, keeping the marking and spacing 

 frequencies equidistant from 1955 cycles. With a swing of ± 55 cycles the 

 distortion was slightly worse than when the swing was ± 35 cycles. The 

 measurements showed least distortion for a swing of ± 15 cycles. It has 

 been previously shown® that the less the swing the smaller the amplitude of 

 the oscillations in the transient for a given channel frequency band width. 

 Accordingly one might expect distortion to be least when the swing is least. 

 If only a small swing is used the signal bias change with carrier frequency 

 drift is worse, unless automatic bias compensation is provided. Greater 

 amplification is also required in the detector-amplifier in order to maintain 

 the same relay operating current. A swing of ± 35 cycles was used in most 

 of the frequency-shift tests as a good compromise between the distortion 

 caused by the greater swings and the severe apparatus requirements and 

 greater susceptibility to noise when using the lesser swings. 



Types of Modulator 



Figure 3 shows distortion characteristics of a frequency-shift arrangement 

 using different types of modulator. Curve A was measured with sinusoidal 

 frequency variation obtained by the use of a diode modulator and low-pass 

 filter at the modulator inputs When the low-pass filter was omitted, the 

 frequency variation was substantially abrupt, and curve B was obtained. 

 When the diode modulator was replaced by a relay modulator, which also 

 produced a substantially abrupt frequency change, curv^e C resulted. There 

 is not much difference between these characteristics at low speeds. At high 

 speeds the abrupt frequency variation appears to give somewhat lower dis- 

 tortion than sinuosidal variation. The distortion shown by curve A depends 

 not only upon the speed and channel filter characteristic but also upon the 

 characteristic of the low-pass filter used in rounding the sent wave in order 

 to produce sinusoidal frequency variation. A considerable amount of care 

 was necessary to prevent this low-pass filter from introducing too much dis- 

 tortion and at the same time to produce sufiicient rounding. The cut-off 

 frequency of this low-pass filter was adjusted at each signaling speed to be 

 about three times the dot frequency. It was apparently low enough to 



