Like Reginald Fessenden, Valdemar Poulsen in Denmark quickly recognized the benefits of continuous-wave (also known as "undamped" and "persistent") radio transmitters, in contrast to the "damped" wave trains produced by the original spark-gap transmitters. To achieve this, in 1902 Poulsen began development of an arc-transmitter, which was much more technically sophisticated than the earlier spark transmitters, and also could be used for audio transmissions.
 
Transactions of the 1904 Saint Louis International Electrical Congress, Volume II, pages 963-971:

SYSTEM  FOR  PRODUCING  CONTINUOUS  ELECTRIC  OSCILLATIONS.
______

BY  V.  POULSEN.
______
figure 1

    It is known that Mr. Duddell, in the year 1900, discovered that a direct-current arc, shunted with a condenser in series with a self-induction, as in Fig. 1, will, under certain conditions, give out a musical note, and transform part of the direct current into alternating current with constant amplitude; the energy dissipated in the condenser circuit in ohmic loss being supplied from the direct current. Duddell found, however, that the arc is only "musical" when the following conditions are satisfied:
    If d V is a small change in the p.d. between the terminals of the arc, and d I the corresponding change in the current through the arc, then dV / dI must be 1), negative; 2), numerically greater than the resistance of the condenser circuit; and 3), numerically less than the resistance of the direct-current circuit in series with the arc. These conditions are fulfilled if the arc is formed between solid carbons.
    This simple way of producing alternating currents of even high frequency seemed, justly, to be the nearly ideal principle for the securing of a system for producing continuous electric oscillations -- i. e., alternating currents of very high frequency.
    When experimenting some years ago with the musical arc, I made an observation which, followed by occasional experiments, led me to the construction of a generator for producing continuous electric oscillations. A short, general description of some of my experiments and arrangements will here be given.
    Fig. 1 shows the diagram of an ordinary musical arc. D is a choking coil, S is the self-induction, and C the capacity in the shunt circuit. R is a regulating resistance inserted in the direct-current circuit. figure 2
    In view of my first experiment, the carbons were placed horizontally and coaxially, as shown in Fig. 2. In this way an ordinary spirit-lamp could be held under the arc in such a manner that the latter and the adjacent parts of the electrodes were quite surrounded by the spirit-vapor. The ammeter A was placed in the direct-current circuit, and the hot-wire ammeter T, in the condenser circuit. The direct current was taken from an ordinary 220-volt lighting circuit.
    When the musical arc was placed in spirit-vapor in the above-mentioned manner, then the note abated in pitch, while at the same time T showed a considerable increase in the alternating current, and the direct current dimished. Furthermore, the emission of light of the arc diminished, as is usual with hydrogen and hydrogen compounds. As with the atmospheric arc, the note also increased in pitch in the alcoholic arc when the direct current was increased. A different length of arc was required in spirit-vapor than in air give a maximum current in the condenser circuit.
    Table I contains the data of two series of experiments; one in which the distance between the electrodes is adapted to spirit-vapor and another in which it is adapted to air.
 
Resistance
of R + D
in ohms.		 Direct-cur-
rent p.d.
between the
electrodes.		 Direct-
current in
amperes.		 Alternating
current in
amperes.		 Distance
between the
electrodes
in mm.		 The arc being in:
54
54
36
36
54
54
36
36		 101
  47
  83
  47
  31
  31
  29
  29		 3.2
3.2
3.8
4.8
3.5
3.5
5.3
5.3		 6.9
2.4
9.0
2.4
3.1
5.8
4.6
6.0		 Abt. 2.0
 "     2.0
 "     2.7
 "     2.7
 "     1.2
 "     1.2
 "     1.0
 "     1.0		 Spirit-vapor.
Air.
Spirit-vapor.
Air.
Air.
Spirit-vapor.
Air.
Spirit-vapor.

    The large choking coil D was without iron and had an inductance of 0.19 henry. The self-induction S without iron was 5.6 X 10-4 henrys and the capacity of the condenser1 was about 2.5 microfarads.
    In the first series of experiments V/I was greater for the spirit arc than for the atmospheric arc. In the second series the pitch of the note was very nearly the same in air and in spirit-vapor; and the inserted resistances and direct currents being the same, the energy in the condenser circuit is proportional to the square of the current. figure 3
    The inserted resistances, as also S and C, were chosen arbitrarily. The superiority of the spirit-vapor became, with relation to the air, greater than in the above-mentioned experiments, when the ratio S/C was taken greater. The same effect as spirit-vapor was given by hydrogen, ordinary coal-gas, and ammonia-gas. As even water-vapor gave an effect similar to that of the hydrocarbons at low frequencies, the effect seemed be produced by the hydrogen. That the effect was not due to the streaming of the gas became evident through experiments in a closed vessel.
    When the image of the hydrogenic arc was projected upon a screen, the condenser circuit not being closed, the arc was observed as a greenish-blue spot with a very faint trace of a purple-colored core. As soon as the arc was made "musical" by closing the condenser circuit, it became thick, the purple-colored core becoming then very marked. When there a copper anode and a carbon cathode were used instead of two carbons, the core was particularly beautiful.
    As I aimed at obtaining frequencies as high as possible, and as the superiority of the hydrogenic arc became more evident when the ratio S/C was made greater, as above mentioned, I laid special stress on dimishing C, which was reduced with good result down to 1 X 10-4 microfarads.
    The hydrogenic arc gave out "musical" notes, or rather electric "notes," of several hundred thousand oscillations per second, and even though of less intensity, some millions of oscillations per second. The excellent resonance effect that can be obtained by these oscillations indicate their continuity, and in the rotating mirror it is seen that the oscillations actually are continuous.
    At high frequency, alcohol did not prove to be as good as hydrogen, coal-gas, or ether. Furthermore, it was shown to be necessary to draw out the arc to a certain length in starting the oscillations. When the oscillations are started, the length, as a rule, can be lessened a little, without the oscillations ceasing. If the length of the arc is increased, then the oscillations continue, and cease only when the distance between the electrodes has become so great that the arc is extinguished. figure 4
    When the arc oscillates in a gas flame, this latter assumes a special form. If the ratio S/C is small, the appearance of the flame and arc is very curious, the gas, or particles in the gas, being projected from the arc with a blowing sound. If the ratio S/C is very small or very great, the arc cannot oscillate.
    On some few occasions I got, when S/C was great, a momentary p.d. between the coatings of the Leyden jar which represented C so great that the edge became luminous and the odor of ozone was present. As this was not repeated later on, I placed my oscillating arc in a transverse magnetic field, under otherwise the same conditions, in order to see whether this would be of avail.
    The increase in the effect was very striking. From the Leyden jar was heard and seen a splendid luminous ring of small discharges from the edge of the inner coating. The heat caused thereby became so great that the tinfoil melted at the edge and the Leyden jar cracked in a circle. When, instead of a permanent magnet, I used an electromagnet, inserted in the direct-current circuit (see Fig. 3), then the arc became more stable and showed, with the electrodes suitably formed, and with even a magnetic field of 1 X 104 to 1:5 X 104 gausses, no tendency to extinguishment.
    In the magnetic field the resistance of the arc, or rather the ratio V/I, is very great, and the more so the greater is S/C. When the electrodes are drawn back from each other in the magnetic field, the direct current decreases until that length of the arc is attained at which the oscillations begin; the direct current then increases again somewhat, that is, the oscillations lessen the resistance of the arc.
    When the hydrogenic arc is placed in a magnetic field as mentioned, S/C can be chosen much greater than otherwise. On the other hand, the magnetic field will do more harm than good when S/C is small.
    The atmospheric musical arc cannot be established with a value of S/C that makes the magnetic field applicable to the hydrogenic arc, and the magnetic field is, moreover, quite inapplicable to the atmospheric musical arc. A magnetic field parallel to the hydrogenic arc shows about the same effect as a transverse field. figure 4a
    As electrodes I have used, besides carbon to carbon, different metals, for example, with good effect + copper cooled by running water, to - carbon. Some forms of water-cooled electrodes are shown in Fig. 4.
    The wear is surprisingly small. Silver, copper, and mercury are about equally good anode metals for the oscillating arc. + copper to - copper proved on some occasions to be of very great effect; but this combination in general gives rise to discontinuities in the oscillations.
    Where there is wanted an arrangement that can stand and take care of itself for a longer time, it is necessary to remove the carbon deposit from the electrodes in order to keep the length of the arc and the shape of the electrodes unaltered. This can be done by scraping the rotating electrodes with knives of hard and fireproof material, such as talc or self-hardening steel.
    If the electrodes are placed horizontally in a transverse magnetic field, then the arrangement ought to be such that the arc is forced upward, at any rate, not downward. An arrangement sufficiently good for many experiments is that shown in Fig. 5. Here a cooled copper anode is fixed opposite to a rotating carbon cathode (speed at the periphery 2 to 5 mm.p.s.) in an ether or gas flame (not a Bunsen burner). The magnetizing coils can replace the choking coil. In order to avoid soot deposits, one may inclose the arc in a case, preferably cooled by water, and let the gas stream out from it, for instance, to a Bunsen burner; if the gas has no outlet, then the effect is lessened gradually, especially, when a strong current is used, the composition of the gas being altered at the same time.
    If an ordinary Leyden jar is used instead of an air or oil condenser, the jar becomes very warm, if the tinfoil does not closely adhere to the glass. figure 5
    In case the coatings closely adhere to the glass everywhere, then the jar can be used, if vaseline or a similar insulating substance is spread over the coatings on the inside and on the outside, so that their edges are well covered by the oil.
    In some recent experiments I obtained, with a frequency of about 5 X 104 in the condenser circuit, about 1560 watts, the arc at the same time taking from the direct-current circuit about 3170 watts; the efficiency was thus about 50 per cent. With 1140 watts in the condenser circuit, the direct watts were 2700, the efficiency thus being 42 per cent. With 464 watts in the condenser circuit, the direct watts were 1070 and the efficiency thus 43 per cent. The frequency being about 1.6 X 105 I had in the condenser circuit 800 watts, the direct watts being 2800; the efficiency thus only proved to be 29 per cent. The supply voltage during all these experiments was 440 volts.
    The supply voltage being 220 volts, I obtained with a frequency of about 5 X 104 in the condenser circuit, 358 watts, the direct watts being 718; this gives an efficiency of 50 per cent. With a frequency of only 3000 to 4000 periods, I had in the condenser circuit 282 watts, the direct watts being 656 and the efficiency thus 43 per cent. figure 6
    In regard to all the above-mentioned experiments no preparations were made to obtain the greatest efficiency or effect. For instance, the arc was placed in a water-cooled vessel, without outlet for the gas, this being necessary to determine the energy; this lessened the intensity of the oscillations, as mentioned above, while at the same time the composition of the gas was altered. At a low frequency the alteration of the gas does not impair the intensity so much, on the other hand, the insulation was very bad in the condenser used for the frequencies of 3000 to 4000.
    That my system for producing continuous electric oscillations admits of handling a good deal of energy at even very high frequency has been proved by different experiments in connection with the ordinary lighting circuit of 220 volts. A resonating coil gave, when it was connected by the Seibt arrangement to an oscillating circuit with the ratio henrys 1 microfarads == 4.8, a noiseless, very warm flame of a length of 12 cm. The frequency was 1.2 X105. If the flame is made short, it can distinctly be seen in the rotating mirror to be continuously oscillating. A large Röntgen tube was placed between the terminals of a coil inductively coupled with an oscillation circuit with a frequency of about 2 X105, and in a short time the cathode and the anti-cathode melted. figure 7
    An ordinary 200-volt incandescent lamp glowed when placed in series with two persons, one of whom was connected with an oscillating circuit, the ratio S/C being about 1. A Seibt resonating coil with the frequency 8.4 X105 gave a flame of a length of about 1 cm; an inductively coupled coil with the frequency 1.1 X106 gave the same length.
    If one surrounds the secondary coil of an ordinary spark coil with windings of thick-copper wire and places these windings in series with a capacity of some microfarads shunting a hydrogenic arc, a very loudly singing flame of a length of 10 to 12 cm or more is obtained. A Röntgen tube with this arrangement gives a very strong radiation.
    When the ratio S/C is great -- about 15 -- there is a considerable p.d. directly between the condenser plates. With a frequency of 50,000 to 150,000 there are thick sparks of 2 to 5 cm long when the self-induction is shunted with a spark-gap.
    Fig. 6 shows a diagram with two oscillating circuits of the same frequency; by means of such an arrangement oscillating flames of about the double voltage can be obtained.
    I noticed that the musical arc placed in nitrogen gave rise to larger alternating currents than in air; at the same time I noticed that the atmospheric arc gave a larger alternating current before the carbons become quite hot. From this I conclude that the musical arc, considered as an electric transformer, is handicapped by the oxygen and that this circumstance is connected with the combustion.
    I could not obtain with the nitrogenic arc as high frequencies, combined with high currents, as with the hydrogen arc.
    Since the oxydation of the electrode material seems to reduce the alternating currents, as above mentioned, it is natural to conclude that the superiority of the hydrogen is partly due to its great affinity for oxygen, which, even in small quantities, must be supposed to affect the oscillations adversely. Without going into a more detailed explanation of the influence of the hydrogen on the musical arc, I will only mention the peculiar position of hydrogen among the elements with respect to velocity of the ions.
    On the basis of the experiments I have made with the "oscillating arc," I believe that it can in future be used as an electric generator for syntonic wireless telegraphy and telephony. Without mentioning other technical uses for which, I believe, it is fitted, I may, finally, express the hope that it will be of a value to physicists and electricians comparable to that of the Rühmkorff coil in the past.
_____________
1. The insulation of the condenser was not good.