Johnson - Tesla Coil Impedance

Johnson - Tesla Coil Impedance
Tesla Coil Impedance
Dr. Gary L. Johnson
Professor Emeritus
Electrical and Computer Engineering Department
Kansas State University
The input impedance of a Tesla coil operated as an ‘extra’ coil, or as a quarter-wave antenna above a ground
plane, is given here. Effects of coil form, wire size, wire
insulation, and humidity are discussed.
va vb
iron core
1. Introduction
A classical Tesla coil contains two stages of voltage
increase. The first is a conventional iron core transformer that steps up the available line voltage to a
voltage in the range of 12 to 50 kV, 60 Hz. The second
is a resonant air core transformer (the Tesla coil itself)
which steps up the voltage to the range of 200 kV to
1 MV. The high voltage output is at a frequency much
higher than 60 Hz, perhaps 500 kHz for the small units
and 80 kHz (or less) for the very large units.
L1 L2
air core
Figure 1: The Classical Tesla Coil
very high. The impedance during conduction depends
on the geometry of the gap and the type of gas (usually
air), and is a nonlinear function of the current density.
This impedance is not negligible. A considerable fraction of the total input power goes into the production
of light, heat, and chemical products at the spark gap.
The lumped circuit model for the classical Tesla coil
is shown in Fig. 1. The primary capacitor C1 is a low
loss ac capacitor, rated at perhaps 20 kV, and often
made from mica or polyethylene. The primary coil L1
is usually made of 4 to 15 turns for the small coils and
1 to 5 turns for the large coils. The secondary coil
L2 consists of perhaps 50 to 400 turns for the large
coils and as many as 400 to 1000 turns for the small
coils. The secondary capacitance C2 is not a discrete
commercial capacitor but rather is the distributed capacitance between the windings of L2 and the voltage
grading structure at the top of the coil (a toroid or
sphere) and ground. This capacitance changes with the
volume charge density around the secondary, increasing somewhat when the sparks start. It also changes
with the surroundings of the coil, increasing as the coil
is moved closer to a metal wall.
The arc in the spark gap is similar to that of an electric arc welder in visual intensity. That is, one should
not stare at the arc because of possible damage to the
eyes. At most displays of classical Tesla coils, the spark
gap makes more noise and produces more light than the
electrical display at the top of the coil.
When the gap is not conducting, the capacitor C1
is being charged in the circuit shown in Fig. 2, where
just the central part of Fig. 1 is shown. The inductive
reactance is much smaller than the capacitive reactance
at 60 Hz, so L1 appears as a short at 60 Hz and the
capacitor is being charged by the iron core transformer
A common type of iron core transformer used for
small Tesla coils is the neon sign transformer (NST).
Secondary ratings are typically 9, 12, or 15 kV and
30 or 60 mA. An NST has a large number of turns
on the secondary and a very high inductance. This
inductance will limit the current into a short circuit at
about the rated value. An operating neon sign has a
low impedance, so current limiting is important to long
The symbol G represents a spark gap, a device which
will arc over at a sufficiently high voltage. The simplest
version is just two metal spheres in air, separated by a
small air gap. It acts as a voltage controlled switch in
this circuit. The open circuit impedance of the gap is
by the symbol M . The coefficient of coupling is well
under unity for an air cored transformer, so the ideal
transformer model used for an iron cored transformer
that electrical engineering students study in the first
course on energy conversion does not apply here.
Figure 2: C1 Being Charged With The Gap Open
transformer life. However, in Tesla coil use, the NST
inductance will resonate with C1 . The NST may supply
two or three times its rated current in this application.
Overloading the NST produces longer sparks, but may
also cause premature failure.
s X
Figure 4: Lumped Circuit Model Of A Tesla Coil, arc
When the voltage across the capacitor and gap
reaches a given value, the gap arcs over, resulting in
the circuit in Fig. 3. We are not interested in efficiency
in this introduction so we will model the arc as a short
circuit. The shorted gap splits the circuit into two
halves, with the iron core transformer operating at 60
Hz and the circuit to the right of the gap operating at
a frequency (or frequencies) determined by C1, L1, L2 ,
and C2 . It should be noted that the output voltage of
the iron core transformer drops to (approximately) zero
while the input voltage remains the same, as long as
the arc exists. The current through the transformer is
limited by the transformer equivalent series impedance
shown as Rs + jXs in Fig. 3. As mentioned, this operating mode is not a problem for the NST. However, the
large Tesla coils use conventional transformers with per
unit impedances in the range of 0.05 to 0.1. A transformer with a per unit impedance of 0.1 will experience
a current of ten times rated while the output is shorted.
Most transformers do not survive very long under such
conditions. The solution is to include additional reactance in the input circuit.
At the time the gap arcs over, all the energy is stored
in C1. As time increases, energy is shared among C1,
L1 , C2, L2 , and M . The total energy in the circuit
decreases with time because of losses in the resistances
R1 and R2. There are four energy storage devices so a
fourth order differential equation must be solved. The
initial conditions are some initial voltage v1 , and i1 =
i2 = v2 = 0. If the arc starts again before all the
energy from the previous arc has been dissipated, then
the initial conditions must be changed appropriately.
With proper design (proper values of C1, L1, C2,
L2 , and M ) it is possible to have all the energy in C1
transferred to the secondary at some time t1 . That
is, at t1 there is no voltage across C1 and no current
through L1 . If the gap can be opened at t1 , then there
is no way for energy to get back into the primary. No
current can flow, so no energy can be stored in L1 , and
without current the capacitor cannot be charged. The
secondary then becomes a separate RLC circuit with
nonzero initial conditions for both C2 and L2, as shown
in Fig. 5. This circuit will then oscillate or “ring” at
a resonant frequency determined by C2 and L2 . With
the gap open, the Tesla coil secondary is simply an
RLC circuit, described in any text on circuit theory.
The output voltage is a damped sinusoid.
Figure 3: Tesla Circuit With Gap Shorted.
The equivalent lumped circuit model of the Tesla coil
while the gap is shorted is shown in Fig. 4. R1 and R2
are the effective resistances of the air cored transformer
primary and secondary, respectively. The mutual inductance between the primary and secondary is shown
Figure 5: Lumped Circuit Model Of A Tesla Coil, arc
above 2 cm. It requires 0.02/80 = 25 µs for the disc
to turn this distance. This time can be shortened by
making the disc larger or by turning it at a higher rate
of speed, but in both cases we worry about the stress
limits of the disc. Nobody wants fragments of a failed
disc flying around the room. The practical lower limit
of arc length seems to be about 10 µs. With larger coils
this may be reasonably close to the optimum value.
Finding a peak value for v2 given some initial value
for v1 thus requires a two step solution process. We
first solve a fourth order differential equation to find
i2 and v2 as a function of time. At some time t1 the
circuit changes to the one shown in Fig. 5, which is
described by a second order differential equation. The
initial conditions are the values of i2 and v2 determined
from the previous solution at time t1 . The resulting
solution then gives the desired peak values for voltage
and current. The process is tedious, but can readily
be done on a computer. It yields some good insights
as to the effects of parameter variation. It helps establish a benchmark for optimum performance and also
helps identify parameter values that are at least of the
correct order of magnitude. However, there are several
limitations to the process which must be kept in mind.
The third reason for concern about the above calculations is that the Tesla coil secondary has features that
cannot be precisely modeled by a lumped circuit. One
such feature is ringing at ‘harmonic’ frequencies. Neither distributed nor lumped models do a particularly
good job of predicting these frequencies. For example,
a medium sized secondary might usually ring down at
160 kHz. Sometimes, however, it will ring down at
3.5(160) = 560 kHz. A third harmonic appears in many
electrical circuits and has plausible explanations. A 3.5
‘harmonic’ is another story entirely.
First, the arc is very difficult to characterize accurately in this model. The equivalent R1 will change,
perhaps by an order of magnitude, with factors like i1 ,
ambient humidity, and the condition, geometry, and
temperature of the electrode materials. This introduces a very significant error into the results.
2. The Extra Coil
As mentioned above, the classical Tesla coil uses two
stages of voltage increase. Some coilers get a third
stage of voltage increase by adding a magnifier coil,
also called an extra coil, to their classical Tesla coil.
This is illustrated in Fig. 6.
Second, the arc is not readily turned off at a precise
instant of time. The space between electrodes must be
cleared of the hot conducting plasma (the current carrying ions and electrons) before the spark gap can return to its open circuit mode. Otherwise, when energy
starts to bounce back from the secondary, a voltage will
appear across the spark gap, and current will start to
flow again, after the optimum time t1 has passed. With
fixed electrodes, the plasma is dissipated by thermal
and chemical processes that require tens of microseconds to function. When we consider that the optimum
t1 may be 2 µs, a problem is obvious. This dissipation
time can be decreased significantly by putting a fan on
the electrodes to blow the plasma away. This also has
the benefit of cooling the electrodes. For more powerful
systems, however, the most common method is a rotating spark gap. A circular disc with several electrodes
mounted on it is driven by a motor. An arc is established when a moving electrode passes by a stationary
electrode, but the arc is immediately stretched out by
the movement of the disc. During the time around a
current zero, the resistance of the arc can increase to
where the arc cannot be reestablished by the following
increase in voltage.
va vb
iron core
L1 L2
air core
C2 @ @ magnifier
Figure 6: The Classical Tesla Coil With Extra Coil
The extra coil and the air core transformer are not
magnetically coupled. The output (top) of the classical coil is electrically connected to the input (bottom)
of the extra coil with a section of copper water pipe
of large enough diameter that corona is not a major
problem. A separation of 2 or 3 meters is typical.
Voltage increase on the extra coil is by transmission
line action (field theory), or by RLC resonance (circuit theory), rather than the transformer action of the
iron core transformer. Voltage increase on the air core
transformer is partly by transformer action and partly
by transmission line action. When optimized for extra coil operation, the air core transformer looks more
like a transformer (greater coupling, shorter secondary)
The rotary spark gap still has limitations on the minimum arc time. Suppose we consider a disc with a radius of 0.2 m and a rotational speed of 400 rad/sec
(slightly above 3600 RPM). The edge of the disc is
moving at a linear velocity of rω = 80 m/s. Suppose
also that an arc cannot be sustained with arc lengths
than when optimized for classical Tesla coil operation.
Although not shown in Fig. 6 the extra coil depends
on ground for the return path of current flow. The
capacitance from each turn of the extra coil and from
the top terminal to ground is necessary for operation.
Impedance matching from the Tesla coil secondary to
the extra coil is necessary for proper operation. If the
extra coil were fabricated with the same size coil form
and wire size as the secondary, the secondary and extra
coil tend to operate as a long secondary, probably with
inferior performance to that of the secondary alone.
There are guidelines for making the coil diameters and
wire sizes different for the two coils, but optimization
seems to require a significant amount of trial and error.
Figure 7: Drive for Tesla Coil
higher order resonances are not exact multiples of the
fundamental frequency. That is, I apply a square wave
of voltage to the feed point, and observe a current that
looks sinusoidal at the fundamental frequency. The
lumped RLC model automatically excludes higher order resonances, so if they are of significance, we must
use a distributed model to describe them. I believe that
higher order resonances are not a problem, at least not
enough of a problem to exclude the use of the lumped
model. The input impedance of the Tesla coil is the
(fundamental) input voltage divided by the current i.
In my quest for a better description of Tesla coil
operation, I decided that the extra coil was the appropriate place to start. It looks like a vertical antenna
above a ground plane, so there is some prior art to draw
from. While the classical Tesla coil makes an excellent
driver to produce long sparks, it is not very good for instrumentation and measurement purposes. There are
just too many variables. The spark gap may be the
best high voltage switch available today, but inability
to start and stop on command, plus heating effects,
make it difficult to use when collecting data.
The lumped model used here is the series resonant
RLC circuit shown in Fig. 8.
I therefore decided to build a solid state driver.
Vacuum tube drivers have been used for many years
and several researchers have developed drivers using power MOSFETs, so this was not entirely new
territory. I used this driver to measure the input
impedance of several coils under various operating conditions, and compared these results with what theory I could find. This paper describes my results.
Some data on my driver can be found on my web site,˜gjohnson.
Figure 8: Tesla Coil with Series Resonant LC Circuit
The resonant frequency is given by
3. The Lumped RLC Model
There are two ways of modeling the Extra Coil: distributed (fields) and lumped (circuits). I spent considerable time with the distributed model, but was unable to predict all the interesting features. The lumped
model is not perfect either, but may be easier to visualize. The system we are attempting to describe is shown
in the next figure.
ωo = √
At resonance, the inductive reactance ωo Lw is canceled by the capacitive reactance −1/ωo Ctc so the current is
The voltage input vi at the base of the Tesla coil may
be either a sine wave or a square wave. The square wave
is composed of an infinite series of cosinusoids, the fundamental and all odd harmonics. The harmonics of the
exciting wave will drive higher order resonances of the
Tesla coil, if these resonances are harmonically related.
It appears, at least for the coils I have built, that any
The magnitude of the voltage across either the inductance or the capacitance is
VC = iXC =
Rtc ωo Ctc
ωo CtcRtc
ceiling) with the coil setting on a large copper sheet.
However, most coils are operated indoors without any
copper sheet. Ground then consists of some combination of a concrete floor, electrical wiring, grounded
light fixtures, and soil moisture. That is, the geometry
necessary even for a numerical solution of capacitance
is difficult to describe precisely.
Books on circuit theory define the quality factor Q
ωo Ltc
ωo CtcRtc
so we can write
VC = Qvi
To obtain the capacitance of a sphere to ground, we
start with the capacitance of a spherical capacitor, two
concentric spheres with radii a and b (b > a) as shown
in Fig. 9.
We see that to get a large voltage on the toroid of
a Tesla coil that it needs to be high Q. We need Rtc
small, Ltc large, and Ctc small. Let us look at ways of
calculating or estimating these quantities.
2a 6
? ?
4. Inductance
Figure 9: Spherical Capacitor
An empirical expression for the low-frequency inductance of a single-layer solenoid is [14, p. 55].
Ltc =
r2 N 2
9r + 10`
It is not practical to actually build capacitors this
way, but the symmetry allows an exact formula for capacitance to be calculated easily. This is done in most
introductory courses of electromagnetic theory. The
capacitance is given by [12, Page 165]
where r is the radius of the coil and ` is its length in
inches. This formula is accurate to within one percent
for ` > 0.8r, that is, if the coil is not too short. It is
known in the Tesla coil community as the Wheeler formula. The structure of a single-layer solenoid is almost
universally used for the extra coil, so this formula is
very important. In normal conditions (no other coils
and no significant amounts of ferromagnetic materials
nearby) it is quite adequate for calculating resonant
1/a − 1/b
If the outer sphere is made larger, the capacitance
decreases, but does not go to zero. In the limit as
b → ∞, the isolated or isotropic capacitance of a sphere
of radius a becomes
The classical Tesla coil has a short primary that is
magnetically coupled into a taller secondary. The paper by Fawzi [5] contains the analytic expressions for
the self and mutual inductances necessary for this case.
A relatively simple numerical integration is required to
get the final values. We can get the same results as
the Wheeler formula by this numerical integration, but
with somewhat less insight as to how inductance varies
with the number of turns, and the length and radius of
the coil.
C∞ = 4πa
Assuming a mean radius of 6371 km, the isotropic
capacitance of the planet earth is 709 µF. A sphere of
radius 0.1 m (a nice size for a small Tesla coil) would
have an isotropic capacitance of 11.1 pF.
The other type of top load used for Tesla coils is
the toroid. The dimensions of a toroid are shown in
Fig. 10.
5. Capacitance
Ctc is much more difficult to calculate or to estimate
with any accuracy. The capacitance value used to determine the resonant frequency of the Tesla coil is a
combination of the capacitance of the coil and the capacitance of the top load, usually a sphere or a toroid,
with respect to ground. As a practical matter, ground
will be different in every Tesla coil installation. Easiest
to model would be an outdoor installation (no walls or
Figure 10: Toroid Dimensions
The analytic expression for the isotropic capacitance
of a toroid involves Legendre functions of the first and
second order. It is much more difficult than the capacitance of a sphere. I discuss this expression at my
web site, and give two corrections to formulas found
in the Moon and Spencer textbook. For our purposes,
empirical formulas for the capacitance of a toroid are
more than adequate. The following are given by [13]
CS =
1.8(D − d)
ln(8(D − d)/d)
CS = 0.37D + 0.23d
(d/D < 0.25)
(d/D > 0.25)
coil near a ground plane, the capacitance increases. As
we add walls and a ceiling, the capacitance increases
some more. The same is true for the toroid. The capacitance increases as it is brought to the vicinity of a
ground plane.
But if we want the isotropic capacitance of the combination of the coil and the toroid (the two items assembled at a remote location), shielding occurs such
that the effective capacitance is less than the sum of
the two isotropic capacitances. We have two opposing trends. The actual isotropic Tesla coil capacitance
is smaller than the sum CM + CS but when the coil
and toroid are brought near grounded surfaces, the effective capacitance to be used in calculating resonant
frequency will increase. The opposing trends suggests
that Ctc might be within 20% of CM + CS for most
Tesla coils. The resonant frequency is related to the
square root of Ctc so a 20% error in capacitance results
in only a 10% error in resonant frequency.
where D is the toroid major diameter, outside to outside, in cm, d is the toroid minor diameter in cm, and
the capacitance is given in pF.
Empirical equations for the isotropic capacitance of
a coil were developed many years ago by Medhurst.
These can be expressed in several different versions, to
meet different needs. The simplest expression for the
isotropic capacitance of a cylindrical coil of wire, with
diameter D and coil length `, is
CM = HD pF
The main benefit of the expressions for CM and CS
is that they allow us to do ‘what if’ analyses relatively
quickly. Questions about the effect of changing coil diameter, coil length, or toroid diameter can be answered
with adequate accuracy.
It is possible to calculate Ctc numerically using
Gauss’s Law. If one is careful about measuring and entering all the dimensions and the locations of grounded
surfaces, one should get a value for Ctc well within 5%
of the correct value. There are programs available in
the Tesla coil community that do this.
where D is in cm, and H is a multiplying factor that
equals 0.51 for `/D = 2, 0.81 for `/D = 5, and varies
linearly between 0.51 and 0.81 for `/D between 2 and
5. Most coilers prefer values for `/D between 3.5 and
4.5, so this linear range is adequate for most purposes.
6. Copper Resistance
An expression for H that works for `/D between 2
and 8 is
H = 0.100976
+ 0.30963
Rtc includes all the different types of losses observed
in an actual Tesla coil, including radiation losses (the
coil putting some of the input power into Hertzian
waves being radiated into space) and dielectric losses
(heat produced in the coil form and the wire insulation). The largest component of Rtc in a well-built coil
is the ‘copper’ resistance Rcu. (Most Tesla coils are
wound with copper wire, but other conductors could be
used as well. The ‘copper’ resistance is the effective resistance of the conducting metal, excluding other types
of losses.) This is found in three steps: First we find
the dc resistance Rdc given the length and diameter of
the wire, and the temperature. We can either look up
the resistance per unit length in a table and multiply
by the length, or we can simply use a ohmmeter.
Another expression for H that works for `/D between 1 and 8 is
` 4
) − 0.0097( )3 + 0.0648( )2
−0.0757( ) + 0.4723
We now have expressions for the isotropic capacitance CS of a toroid (or sphere) and the isotropic capacitance CM of a coil. We want to somehow use these
expressions to find the effective capacitance Ctc for the
Tesla coil with the toroid (or sphere) on top the coil.
Unfortunately, life is not that simple. As we bring the
Second, we find the ac resistance of the same wire in
one single straight length (not in a coil). Rac is greater
than Rdc because of the skin effect that causes less of
the total current to flow in the center of the wire. We
can write
Rac = KskinRdc
and z1 is the center-to-center spacing between adjacent turns, all in consistent units. This table indicates
that the proximity effect can easily double or triple the
measured input resistance over that predicted by Rac
for a straight wire of the same length.
where the multiplying factor due to skin effect is
greater than unity and less than perhaps three. The
general procedure for finding this multiplying factor is
found in many electromagnetic theory books.
We now need to return to a discussion of the skin
effect. An expression for skin depth can be derived as
Third, we find the ac resistance of the wire when
formed into a coil. The resistance is increased because of the proximity effect. I have not found a treatment of the proximity effect for a solenoid in a textbook. I found two papers that deal with this effect,
by Medhurst [10] (of Medhurst capacitance fame), and
Fraga [6]. Neither paper deals with the typical values
of skin depth δ versus wire diameter found in Tesla
coils. Medhurst looked at the high frequency case
where b δ while Fraga looked at the low frequency
case (b ≤ δ). Tesla coils are usually operated at frequencies where the wire radius will be between one and
five skin depths.
δ= √
Rcu,F = KF Rac
The skin depth for copper at 20o C is
δ= √
where b is the radius of the wire and δ is the skin depth,
in consistent units. This equation is only valid for δ b. As might be expected, this excludes most Tesla coils,
so we must find other expressions.
Rac = Rdc
A typical analytic approach is to start with
Maxwell’s Equations and write a differential equation
for the current density inside the wire. The solution of
this differential equation is a Bessel function. Things
start to get tedious as one tries to keep track of the
real and imaginary parts of the Bessel function (the
ber and bei functions). I show some of the details at
my web site.
Table 1: Experimental values of KM and KF (last column), the ratio of high-frequency coil resistance to the
resistance at the same frequency of the same length of
straight wire.
Most introductory electromagnetic theory books derive the expression for ac resistance as
where KM and KF are looked up in Table 1.
`w /D
3.54 3.31
3.05 2.92
2.60 2.60
2.27 2.29
2.01 2.03
1.78 1.80
1.54 1.56
1.32 1.34
where f is the frequency in Hz, µ is the permeability of the conductor (4π × 10−7 for nonferromagnetic
materials), and σ is the conductivity.
We might define a Medhurst copper resistance Rcu,M
and a Fraga copper resistance Rcu,F from these two
papers where
Rcu,M = KM Rac
The approximations developed by Terman [14], who
has a detailed discussion of this topic, will be quite
sufficient for our purposes. He defines Rac in terms of
a parameter x, where
x = πd
for nonmagnetic materials. Here, d is the conductor
diameter in meters, f is the frequency in Hz, and ρ is
the resistivity in ohm meters. As x gets very small,
due to either low frequency or small wire, the ac resistance approaches the dc resistance. Above about
x = 3, Rac /Rdc varies essentially linearly with x according to the expression
In this table, `w is the coil winding length, D is the
coil diameter, d is the diameter of the copper wire,
= 0.3535x + 0.264
64.08 mils or 1.628 mm, and the number of turns
was 387. The center-to-center spacing between adjacent turns z1 was z1 = `w /(N − 1) = 3.02 mm.
The quantity d/z1 = 1.628/3.02 = 0.54. The quantity `w /D = 1.166/0.396 = 2.94. By interpolation
in Table 1, KM is found to be 1.85. Therefore, the
predicted ohmic resistance Rcu,M of the coil would
be (1.85)(10.85) = 20 Ω (at sufficiently high frequencies). Interpolation in the last column of Table 1 gives
KF = 2.348 and Rcu,F = (2.348)(10.85) = 25.5Ω. The
measured input resistance is about 23 Ω at resonance.
This measured resistance includes dielectric losses, radiation losses, etc. in addition to ohmic losses. The
Medhurst estimate Rcu,M allows 3 Ω for these losses,
which is probably not far from reality. The Fraga estimate Rcu,F is at least 10% too high, and perhaps 20%
too high for this particular coil.
(x > 3) (21)
Terman gives the following tabular values of Rac/Rdc
for x between 0 and 3.
Table 2: Rac/Rdc for
0.5 1.0003
0.6 1.0007
0.7 1.0012
0.8 1.0021
0.9 1.0034
1.0 1.005
1.1 1.008
1.2 1.011
1.3 1.015
1.5 1.020
various values of x.
Rac /Rdc
1.5 1.026
1.6 1.033
1.7 1.042
1.8 1.052
1.9 1.064
2.0 1.078
2.2 1.111
2.4 1.152
2.6 1.201
2.8 1.256
3.0 1.318
The Medhurst estimate appears to be better for some
coils, while the Fraga estimate seems better for other
coils. I tested nine different coils, eight with three different top loads to get three different frequencies, so I
had a total of 25 cases where I could compare the calculated Rcu,M and Rcu,F with the measured Rtc. Rcu,M
was less than Rtc by a ‘reasonable’ amount 15 times,
while Rcu,F was less than Rtc by a ‘reasonable’ amount
11 times. When both calculated values were greater
than the measured, Rcu,F was closer to the measured
case twice, while Rcu,M was closer three times. There
were six cases (all three frequencies for two short, fat
coils) where RF was less than Rtc by an ‘unreasonable’
amount. One example was a coil wound with 16 gauge
wire on a barrel where I measured Rtc = 93.1Ω, and
calculated Rcu,M = 80.9Ω and Rcu,F = 51.1Ω. The
conclusion seems to be that as long as we stay away
from short, fat coils, either Medhurst or Fraga estimates will be in the right ball park.
For copper, ρ = 1.724 × 10−8 ohm meters. For those
of us still using wire tables in English units, where wire
diameters are given in mils (1 mil = 0.001 inch), Terman [14] has reduced the expression for x to
x = 0.271dm
fM Hz
where dm is the wire diameter in mils and fM Hz is the
frequency in MHz.
For example, I used 482 meters of 14 ga copper wire
to wind a coil. The dc resistance is 3.99 Ω at 20o C. The
nominal diameter of 14 ga wire is 64.08 mils. Assume
the resonant frequency to be 160 kHz. We calculate x
Table 3 illustrates a common but non-intuitive observation in the Tesla coil community, that coils with
a smaller wire size (higher resistance per unit length)
may work as well as coils wound with larger wire. We
see this as we compare coils 18T and 22T. Coil 18T is
tight wound with 18 gauge Essex copper magnet wire,
coated with what they call Heavy Soderon. Coil 22T is
tight wound with 22 gauge tinned copper hook-up wire,
with what I assume to be PVC insulation. The thicker
insulation means the copper of each turn has a greater
spacing, which reduces the proximity effect. The Medhurst factor KM drops from 3.15 to 1.62 between the
two coils, due mostly to this greater spacing.
x = 0.271(64.08) 0.16 = 6.946
The ac resistance of the wire in the coil (assuming
the wire is uncoiled and is supported in one straight
line) is then
Rac = 3.99(0.3535(6.964) + 0.264) = 10.85 Ω
We see that skin effect makes a significant difference in resistance, especially where larger wire sizes or
higher frequencies are used.
The dc resistance of coil 22T is twice that of coil 18T,
so the natural assumption would be that Rtc would be
larger for coil 22T. However, the thinner wire is more
The diameter D of this coil was 0.396 m, the winding length `w was 1.166 m, the wire diameter d was
We are now ready to discuss (at least qualitatively)
other losses besides those due to copper resistance.
These are illustrated in the next figure.
Table 3: Predicted and Measured Coil Resistance for
Several Coils
d, mm
D, meters
`w , meters
wire, meters
turns N
Ltc , mH
CM , pF
f0 , kHz
Figure 11: Detailed Lumped Model of Tesla Coil
Rrad refers to the power radiated away into space
as an electromagnetic wave. In all my tests, Rrad was
very close to zero, probably not larger than 0.01 Ω. I
was unable to detect a signal from the coil more than
perhaps 100 m away, using a high quality short wave
receiver that functioned down to 150 kHz. A Tesla coil
is a really bad antenna.
Reddy refers to eddy current losses in the toroid (or
sphere) on top the coil, and in any other conducting
surfaces nearby. Eddy current losses can be substantial
in some engineering applications. A power frequency
transformer built of solid steel rather than thin steel
laminations would melt quickly! A spun aluminum
toroid exposed to the magnetic field of the Tesla coil
current will certainly have eddy currents. There has
been more than one serious Tesla coiler who has been
tempted to cut radial slots into the toroid to reduce
these eddy currents.
effective in its use of available cross section. Rac = 24Ω
for coil 18T at 236.9 kHz, and 35.8 Ω for coil 22T at
301.5 kHz, a difference of 49% rather than 100%. Then
when we multiply by KM , we get Rcu,M = 76.50Ω for
coil 18T, and 58.38 Ω for coil 22T. Coil 22T has 200%
of the dc resistance of coil 18T, but only 76% of the
effective resistance during operation at frequency f0 .
The proximity effect increases the resistance of both
coils above the ac resistance value, but far more for
the tighter effective spacing of coil 18T, so coil 22T is
actually the superior coil for Tesla coil activity. The
same observation holds as larger top loads are used,
driving the resonant frequency down.
I built a toroid of 0.25 inch copper tubing pieces on
insulating disks, connected together at one point by a
conducting disk. The ends were placed into heat shrink
tubing, which was then shrunk to hold the ends a fixed
small distance apart. This toroid was then compared
with a spun aluminum toroid of similar capacitance,
and also with a smaller toroid made of one inch copper tubing with diameter slightly greater than that of
the coil form. The smaller toroid was an attempt to
get a shorted turn as near to the coil as possible. It
lacked the capacitance to be an effective toroid for long
sparks, of course. I could not find any significant difference in input impedance between the insulated toroid
and the spun aluminum toroid. The shorted copper
ring, however, had about 10% higher input impedance
than the toroid that was not a shorted turn. This suggests that you would not notice any improvement if you
cut your beautiful spun aluminum toroid into pieces to
eliminate eddy currents. The effect is there, and can
be measured if one really works at it, but is not that
significant in most situations.
The measured values for Rtc are even less than the
Medhurst predicted values, 11 Ω less for coil 22T and 6
Ω less for 18T at frequency f0 . It is always possible that
my experimental technique was lacking on these two
tests. I prefer to think, however, that the assumptions
made by Medhurst do not match reality for these two
coils, so that the ‘correct’ proximity factors would be
a little less than shown in Table 1 for these particular
coil parameters.
7. Other Losses
formance. In cases where moisture is a factor, performance might improve after a period of operation which
caused the coil form to heat up and dry out.
Overall, my tests indicated that Reddy is no more
than a few percent of RT C . If a little thought is given to
separation of conducting materials from the immediate
vicinity of the coil, eddy current losses can be ignored.
Effects of humidity are shown in two sets of input
impedance data in Table 4 for 3/17/01 and 4/6/01.
The coils were located in the bay of a large Morton
building (a metal skin building, not heated or air conditioned, that might be used for storage of tractors and
other agricultural equipment). The Tesla coil driver
and other test equipment were located in a climatecontrolled room built into a corner of the Morton building. The 3/17/01 data were collected when the bay
temperature was about 9o C and the relative humidity
was about 28%. On 4/6/01 the temperature was about
17o C and the relative humidity was 100%. It had been
damp all week with heavy fog the day before. It would
be rare for a Tesla coil to be operated with significantly
lower relative humidity or total moisture in or on the
dielectrics than the conditions of 3/17/01, and likewise
for more moisture than 4/6/01.
The model indicates that when a spark occurs, the
equivalent resistance Rspark increases from zero to
some finite value, so the input resistance increases during a spark. This is exactly what happens experimentally. The input current drops when the spark starts,
for a constant supply voltage.
The copper, eddy current, and radiation losses are
all proportional to the square of the current in the coil.
The losses in the hot plasma of the spark are such that
the spark losses may not be proportional precisely to
the square of the current, but are definitely related to
some function of the current. On the other hand, the
dielectric losses are proportional to the square of the
voltage across the coil. I try to show that with resistors across the capacitance CM . There are two distinct
dielectrics, the coil form and the wire insulation, represented by Rcf and Rwi. Air forms a third dielectric,
but dry air is basically lossless. Humidity in the air
does add losses, but this humidity soaks into the coil
form and wire insulation, increasing the losses there. I
did not try to distinguish between losses in humid air
and losses of the coil and coil form.
Table 4: Rtc Measured on Two Different Days
For single-frequency, steady-state operation, the parallel combination of a capacitor and two resistors can
be modeled as a series capacitor and resistor, call it
Rdie. This is straightforward Circuit Theory I, but a
bit tedious. We write an expression for the parallel
impedance of Rcf , Rwi, and the capacitance, rationalize it, and simplify the real term. We assume that the
parallel resistances are much larger than the capacitive
reactance, as they will be for any coil with acceptable
losses. Calculation details can be found at my web site.
We see that two of the coils experienced large
changes in Rtc, coils 16B and 22B. Both coils used
a plastic barrel as a coil form that I thought was
polyethylene. I got the barrels at the local recycling
plant. Coil 22B used the same type of wire as coil 22T
which was wound on a piece of PVC, so the difference
in Rtc between these two coils had to be the coil form.
These results indicate that some coil forms are worse
than others. These barrels evidently soak up water in
amounts sufficient to raise the input impedance by a
factor of two.
One line of analysis shows that Rdie
√ varies as 1/f .
Generally speaking, Rac increases as f, and Reddy increases as f 2 . Rrad will increase at a rate somewhere
between f and f 2 . Rspark can be ignored below the
spark inception voltage. Depending on which terms
are dominant loss terms, we may not see a pronounced
change in input impedance with frequency. That has
been my experience. Input impedance will drift from
day to day, (mostly with humidity), but there is no
obvious frequency dependence. Of course, other things
are happening. We know that Rac increases with temperature, while Rdie increases with humidity. If these
were the only factors, we would expect a cold, dry winter day to have the lowest impedance, and a hot, muggy
day to have the highest impedance and the worst per-
Coils 14T, 18T, 20T, and 22T were wound on PVC
while coils 14S and 18B were wound on polyethylene.
Only coil 20T had any type of coating put on top the
winding (polyurethane). Both PVC and polyethylene
appear to have about the same increase in Rtc with
humidity, so it is hard to argue that one should spend
more money on the more expensive polyethylene.
Figure 12: Square wave of voltage, sine wave of current at base of coil 12T below breakout.
8. Impedance During Sparking
As long as one stays with good quality coil forms,
it appears that high humidity will raise Rtc by 5–10%
from the low humidity case. This variation of Rtc with
humidity makes it pointless to do more analytical or
computational work to get better estimates of Rcu,M
or Rcu,F , since we appear to be within the 5–10% range
of accuracy for most cases already.
I calculated the input impedance of the Tesla coil
from measured values of vi and i1 , at or very near resonance. I would apply a square wave voltage from my
driver and measure it and the resulting current with a
HP54645D oscilloscope. If the voltage was below that
necessary for a spark to occur, current would rise to
a steady-state value. Steady-state waveforms are illustrated in Figure 12.
In my opinion, a complete distributed model will not
be any better in dealing with skin effect, proximity effect, and dielectric losses, and would certainly be more
of a programming problem. The one thing that this
lumped model cannot deal with directly is the displacement current effect. In a lumped model, the current
must be the same at all points of the circuit. Actually the current varies from one point in the coil to
another point, due to capacitive and inductive effects.
It is counter-intuitive, but the current may actually increase from the base feed point to about a third of the
way up the coil before starting a monotonic decrease to
a minimum at the top of the coil. A distributed model
can determine the actual current distribution, which
can then be used to find a predicted effective resistance
of the coil. The computational effort is much greater
than the table look-up techniques used to get Rcu,M
and Rcu,F for the lumped case. Whether the accuracy
of the predicted resistance is sufficiently improved to
justify the extra effort is open to question.
Voltage is actually the voltage output of a 10:1 voltage divider. The vertical scale for channel A1 is thus
500 V/div rather than 50. At the lower left of the
screen image we see Vrms(A1)=44.34 V. The rms value
would actually be 443.4 V. The rms of a perfect square
wave is the same as the peak value, so I am applying a
voltage of approximately ±443.4 V to the coil.
The current waveform is the voltage across a 20 mΩ
resistor. The current corresponding to a voltage of
293.3 mV is
i1 =
= 14.67
The combination of a square wave voltage and a sinusoidal current is different from what we are used to,
so we need to go back to circuit theory to make sure
we have all the correct multiplying factors.
increase slows as losses increase. Current will hit its
limit when the input power is equal to the losses, in
this case between 2 and 3 ms after start.
Suppose that we apply a sinusoidal voltage Vp sin ωt
to a non-inductive resistor. The resulting current is
Ip sin ωt. The average power is
Pave =
As current increases, the voltage drop across the
MOSFETs/IGBTs and the droop in the electrolytic
capacitors becomes greater. The decrease in voltage
applied to the Tesla coil is not great, but is most easily noticed when the gate drives are turned off at the
2.1 ms point. Energy stored in the Tesla coil has to
be dissipated somehow. Instead of power flowing from
inverter to coil, it now flows in the opposite direction.
Voltage is constrained by the built-in diodes of the IGBTs. Power supply capacitors are now being charged
instead of discharged. All the voltage drops in wiring
and the IGBTs reverse in sign. We therefore see a small
step increase in voltage when the gate drive signals are
Vp Ip
Vp Ip
= Vac Iac
Vp Ip sin2 θdθ = √ √ =
2 2
When the square wave voltage produces a sinusoidal
current, the integral for average power becomes
1 π
Vp Ip sin θdθ = Vp Ip
π 0
Vp ( 2Iac ) = 0.9VacIac
For this case, the average power is no longer the simple product of rms voltage and rms current (as for dc
and single frequency sinusoids), but has a 0.9 multiplying factor. The difference is due to the fact that the
square wave voltage is composed of an infinite series of
harmonics (fundamental, third, fifth, etc.). Each harmonic contributes to the rms value of the square wave.
The current has no harmonics, so the higher voltage
harmonics do not produce any contribution to the apparent power.
About 0.6 ms after gate drive turn-off, the stored
energy is no longer able to force the IGBT diodes into
forward conduction. Without a power supply affecting the circuit, we now have a classic RLC ring down.
Both voltage and current are decreasing exponentially.
During this portion of the cycle, the IGBTs are acting
as capacitors.
When the supply voltage is increased, a spark occurs when the toroid reaches a sufficient voltage. Figure 14 shows the current building up slowly until spark
initiation at about 1.4 ms and then a more rapid decline. The jagged appearance of the wave envelope is
due to the scope’s algorithm to select representative
points from the long waveform data set and does not
imply that the current is experiencing rapid excursions.
When one spreads out the waveform, it can be seen
that the current looks as well behaved as the current
in Figure 12 except that it grows by a small amount
each cycle until breakout.
Actually the product 0.9VacIac is the apparent power
S for these two waveforms, which happens to be equal
to the average power when the waves are in phase. If
there is a phase shift, Pave will be less than S by a
factor cos θ, where θ is the phase angle between the
two waves. I can readily determine S from the product
of rms voltage and rms current (and a 0.9 multiplying
constant). I always try to operate with voltage and
current in phase, so S can be used as an approximation
for Pave.
One cycle at 176.7 kHz lasts for about 5.66 µs. If the
spark starts at 1400 µs, then it took 1400/5.66 = 247
cycles to get to breakout. Just before breakout, the
voltage was about 604 V and the current was 17.1 A. I
had to expand the waveforms near the 1400 µs point to
get these values. The instantaneous power (the average
power at that instant) was about 9.3 kW. The current
during the spark was about 4 A.
The average power being applied to the coil in Fig. 12
is approximately
Pave = 0.9VacIac = (0.9)(443.4)(14.67) = 5854
Fig. 13 shows the voltage applied to another Tesla
coil and the resulting current for about 4.5 ms. The
voltage is A1 at the top of the figure and the current is
A2 at the bottom. Operation is well below breakout.
The driver is supplying voltage from storage capacitors
for about 2.1 ms. Current builds up as predicted by either transmission line theory or lumped RLC analysis.
Current builds nearly linearly at first, then the rate of
I increased the voltage about 50% and got the current waveform in Figure 15. The spark now starts at
about 0.93 ms. Just before breakout, the voltage was
about 914 V, the current was 21.5 A, and the instantaneous power was about 17.7 kW. The current during
the spark remained at about 4 A.
Figure 13: Tesla Coil Voltage and Current Waveforms
particle from a radioactive decay. At higher levels of
voltage, one gets a streamer each pulse train, but the
time from start to discharge, and the resultant power
input just before discharge, can still easily vary by 10%.
The coil appears to operate as a constant-current
device in breakout. Both the input dc current and the
ac current into the coil stay about constant, perhaps
even decreasing a little, as the supply voltage is increased. The input impedance increases proportional
to the input voltage, in order for the current to remain
constant. This is somehow caused by the dynamics
of the hot plasma in the discharge. I do not have a
detailed explanation.
If power continues to be supplied to the discharge
after the onset, the discharge does not get longer, but
instead gets ‘fatter’. Figure 16 shows two streamers
where the driving voltage was removed soon after the
streamers started. The streamers have a thin, violet, anemic appearance. Figure 17 shows two different streamers where the driving voltage was left in
place for perhaps 2 ms after the onset. The streamers now appear richer, fuller, and whiter. The trend
continues with even longer applied voltage as shown in
Fig. fig:fat.
The ‘inertia’ of the electrons and ions on and around
the toroid play an important part in the length of the
streamer. Increasing the supply voltage causes the current and the energy stored in the coil to build up more
rapidly. This more rapid growth allows less time for
the breakdown of air to occur, with the result of more
energy stored in the coil when breakdown does occur.
This greater energy translates to a longer streamer.
The camera is able to show details that the eye
cannot follow. The streamer does not get uniformly
thicker, but rather shows charge leaving the main discharge path at all the ‘corners’ of the original discharge
path and then curving back toward the main path. It
is almost as though the charge carriers in the discharge
have inertia, such that they want to continue traveling
in a straight line when a ‘corner’ is encountered. If the
only forces involved were the Coulombic forces, the es-
The nature of the discharge process causes variability in the exact amount of current necessary to produce
a streamer for any given shot. At lower levels of voltage, a streamer might be produced one time out of ten
applied pulse trains. The time lag between start and
discharge might vary by a factor of two, probably depending on the chance arrival of a gamma ray or a beta
Figure 14: Current Waveform with Voltage = ±604 V
remains valid at higher powers, an input power of 1
MW would produce a spark length of 14 feet. If we
use IGBTs that can safely deliver 50 A at Tesla coil
frequencies, this requires a supply voltage of ±20, 000
V. If the individual IGBTs have a working voltage of
1000 V, then a chain of 20 IGBTs would be necessary
for each side of the power supply. I will not predict that
such a driver cannot or will not be built. I will just say
that getting long sparks from a solid state driver is not
caping charges would not be drawn back to the main
path. But there are also Amperian forces. Two parallel currents experience an attractive force. Evidently
these Amperian forces are stronger than the Coulombic forces in this case. I have not made any attempt to
quantify these forces.
A casual examination of these (and many other) photos suggests that the escaping charges travel in a spiral around the main path before returning to it. This
might be an optical illusion. It would require two or
more cameras orthogonal to the streamer and each
other to prove or disprove this appearance. This would
be something to check in any further research.
I put a bump on the toroid so all sparks started from
the same point, and placed a sheet of plastic with concentric circles behind the bump. I then set the driver
to produce a few sparks per second and observed the
length of the discharges to air. I would also determine
the average power input just before discharge. Then I
would increase the voltage and repeat the process. I
found the spark length to be related to the square root
of power according to the formula
[1] Corum, J. F. and K. L. Corum, “A Technical
Analysis of the Extra Coil as a Slow Wave Helical Resonator”, Proceedings of the 1986 International Tesla Symposium, Colorado Springs, Colorado, July 1986, published by the International
Tesla Society, pp. 2-1 to 2-24.
[2] Corum, James, F., Daniel J. Edwards, and Kenneth L. Corum, TCTUTOR - A Personal Computer Analysis of Spark Gap Tesla Coils, Published by Corum and Associates, Inc., 8551 State
Route 534, Windsor, Ohio, 44099, 1988.
where P is measured in watts. For the current waveforms of Figure 14 and Figure 15 the corresponding
spark lengths were 16.4 and 22.6 inches. If the formula
[3] Couture, J. H., JHC Tesla Handbook, JHC Engineering Co., 19823 New Salem Point, San Diego,
CA, 92126, (1988).
`s = 0.17 P
Figure 15: Current Waveform with Voltage = ±914 V
Figure 17: Onset plus 2 ms
Figure 16: Onset plus 0.5 ms
2, March/April 1978, pp. 464-468.
[4] Cox, D. C., Modern Resonance Transformer Design Theory, Tesla Book Company, P. O. Box
1649, Greenville, TX 75401, (1984).
[6] Fraga, E., C. Prados, and D.-X. Chen, “Practical
Model and Calculation of AC Resistance of Long
Solenoids”, IEEE Transactions on Magnetics, Vol.
34, No. 1, January, 1998, pp. 205–212.
[5] Fawzi, Tharwat H. and P. E. Burke, The Accurate Computation of Self and Mutual Inductances of Circular Coils, IEEE Transactions on
Power Apparatus and Systems, Vol. PAS-97, No.
[7] Hull, Richard L., “The Tesla Coil Builder’s Guide
to The Colorado Springs Notes of Nikola Tesla”,
[11] Peterson, Gary L., “Project Tesla Evaluated”,
Power and Resonance, The International Tesla
Society’s Journal, Volume 6, No. 1, January/February/ March 1990, pp. 25-34.
[12] Plonus, Martin A., Applied Electromagnetics,
McGraw-Hill, New York, 1978.
[13] Schoessow, Michael, TCBA News, Vol. 6, No. 2,
April/May/June 1987, pp. 12-15.
[14] Terman, Frederick Emmons, Radio Engineers
Handbook, McGraw-Hill, 1943.
[15] Tesla, Nikola, Colorado Springs Notes, A. Marincic, Editor, Nolit, Beograd, Yugoslavia, 1978, 478
Figure 18: Onset plus 4 ms
Tesla Coil Builders of Richmond, 1993.
[8] Johnson, Gary L., “Using Power MOSFETs To
Drive Resonant Transformers”, Tesla 88, International Tesla Society, Inc., 330-A West Uintah,
Suite 215, Colorado Springs, CO 80905, Vol. 4,
No. 6, November/December 1988, pp. 7-13.
[9] Johnson, Gary L., The Search For A New Energy
Source, Johnson Energy Corporation, P.O. Box
1032, Manhattan, KS 66505, 1997.
[10] Medhurst, R. G., “H.F. Resistance and SelfCapacitance of Single-Layer Solenoids”, Wireless
Engineer, February, 1947, pp. 35-43, and March,
1947, pp. 80-92.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF