1

The RMS Average.

By David W. Knight
Version 1.00. 11
th
November 2009.
D. W. Knight, 2009.
David Knight asserts the right to be recognised as the author oI this work.
The most recent version oI this document can be obtained Irom the author's website:
http://www.g3ynh.inIo/
Introduction
Most people who work with electricity are aware that the RMS oI a sinusoidal voltage or current
waveIorm is 1/\2 times the peak value. AIter that however, knowledge tends to become a little
hazy. Design and measurement errors due to an incorrect understanding oI the RMS can lead to
expensive and sometimes dangerous Iailures in power supplies, ampliIiers and transmission
equipment. This article attempts to clariIy the issue.
The heating effect of a sinusoidal voltage or current
Let us begin by considering a pure sinusoidal voltage. The RMS average oI the wave is the
equivalent DC voltage which, when connected across a resistance (R say), will produce the same
heating eIIect. The heating eIIect is the average energy delivered per unit-oI-time, i.e., the average
power, which is given by:
P
av
V
rms
/ R . . . . (1)
The alternating voltage is described by the expression:
V V
pk
Sin
Where V
pk
is the peak or maximum instantaneous voltage magnitude, and 2aIt is the time-
varying phase angle. II the voltage is applied to a resistor, the instantaneous power will be:
P V / R
i.e.:
P V
pk
Sin / R
but, using the standard trigonometric identities:
Sin 1 - Cos
and
Cos(2) 2Cos - 1
i.e.,
Cos | 1 Cos(2)| / 2
we obtain the substitution:
Sin 1 - | 1 Cos(2) | / 2 | 1 - Cos(2) | / 2
Hence the instantaneous power Iunction is:
P V
pk
| 1 - Cos(2) | / ( 2 R )
This is a sinusoid at twice the Irequency oI the voltage wave, with an oIIset so that it just touches
zero and is always positive. The average oI a pure sinusoid is zero, and so the average power is
obtained simply by deleting the cosine term:
P
av
V
pk
/ ( 2 R )
comparing this with (1) we have:
2
V
rms
/ R V
pk
/ ( 2 R )
i.e.
V
rms
V
pk
/ \2
Also, by a similar argument using
P
av
I
rms
R
it can easily be shown that
I
rms
I
pk
/ \2
The square-Root of the Mean of the Squares
Finding the average power delivered by a pure sinusoidal voltage or current is easy; but we also
need a general method which works Ior any waveIorm. Without it we cannot even calculate the
power ratings Ior the resistors in simple narrow-band ampliIiers, because the combination oI a
direct and a sinusoidal current is no longer purely sinusoidal. It turns out that the required technique
is analogous to the method used to calculate the standard deviation oI a set oI measurements.
First square the voltage (or current) waveIorm to obtain a time-varying Iunction proportional to
the instantaneous power. Now sample that power Iunction at a large number oI equally spaced
intervals over a whole cycle. Add the sample values together and divide by the total number oI
samples. This gives a quantity proportional to the mean (i.e., average) power. Now take the square
root to get back to the original units oI voltage (or current). The voltage (or current) obtained is the
square-Root oI the Mean oI the Squares, commonly known as the RMS.
To Iind the RMS with perIect accuracy, oI course, we need to take an inIinite number oI samples.
The summation used to obtain the average thus becomes an integral. We need to take the average
over the repetition interval oI the waveIorm, so iI we use as our relative measure oI time, we need
to integrate over the interval between 0 and 2a radians (i.e., over 360). The interval between
samples is the inIinitesimal d, and so the (inIinite) total number oI samples is deIined as 2a/d.
Hence, dividing the inIinite sum by the inIinite total number oI samples, we have:
V
rms

1
2a
2a
}
0
V d (2)
OI course, this doesn't look quite as easy as the rule we know Ior dealing with pure sine waves, but
this will not prove to be a serious problem. As anyone who has ever used a radio receiver ought to
have noticed, and J B J Fourier discovered long beIore spectrum analysers were invented, an
arbitrary waveIorm can always be represented as a sum oI sinusoids plus a constant. Squaring a
series can be a messy business, but it turns out that the cross-terms generated by squaring a Fourier
series cancel in the integration process. The physical reason is that a resistive load is a linear
network. The various components oI the signal do not mix, and so can be considered to exist
independently.
3
The RMS of a pure sine wave
BeIore complicating matters, let us Iirst check that the general method really works Ior the case Ior
which we already know the answer. A pure sinusoidal voltage is given by the expression:
V V
pk
Sin
Substituting this into (2) gives:
V
rms

1
2a
2a
}
0
V
pk
Sin d
Earlier on we obtained the identity:
Sin | 1 - Cos(2) | / 2
and so we have:
V
rms

V
pk

4a
2a
}
0
| 1 - Cos(2) | d
Now, as any good book oI standard integrals will tell you:
}Cos(ax)dx (1/a) Sin(ax)
and so the solution is:
V
rms

V
pk

4a

[
[

- () Sin(2)

[
[

2a
0
i.e.:
V
rms

V
pk

4a
| (2a - 0) - (0 - 0) |
Hence:
V
rms
V
pk
/ \2
The RMS of an offset sine wave
Now let us consider the case when a sinusoidally varying voltage is added to a constant (i.e., DC)
voltage. In this case the instantaneous voltage can be written:
V V
dc
V
pk
Sin
where V
pk
is deIined as the peak value oI the sinusoidal component on its own, not the peak value
oI the total. Substituting this into (2) gives:
V
rms

1
2a
2a
}
0
( V
dc
V
pk
Sin ) d
4
Which can be multiplied out:
V
rms

1
2a
2a
}
0
( V
dc
2 V
dc
V
pk
Sin V
pk
Sin ) d
Now we have produced a cross-term; but since the phrase "linear network" has already cropped-up,
we must suspect that there will be no interaction between the components in the Iinal reckoning. So,
oI course, we notice that the cross-term is a sinusoid, and all sinusoids average to zero over an
integer number oI cycles. Hence, provided that the integration is carried out over the repetition
interval oI the waveIorm, any Iirst-order (i.e., not squared) sine and cosine terms can simply be
deleted. The mathematical reason is that, Ior integer n:
Sin(2na)0 , Sin(0)0 and Cos(2na)Cos(0)1
Hence, deleting the Iirst-order sine term, and substituting Ior Sin as beIore:
V
rms

1
2a
2a
}
0
V
dc
V
pk
() |1-Cos(2)| } d
Now the Cos(2) term dies out in the averaging process, and the solution oI the integral is simply a
matter oI multiplying the remaining terms by a Iactor oI 2a, which cancels with the 1/2a beIore the
integral symbol. Hence:
V
rms
V
dc
V
pk
/2
But now let us deIine a voltage V
ac
, which is the RMS value oI the AC component on its own. Then,
using the deIinition Ior the RMS oI a pure sinusoid, we have:
V
ac
V
pk
/2
Hence:
V
rms
V
dc
V
ac

When an alternating voltage (or current) with a DC oIIset is applied to a resistance, the total power
is given by calculating the power Ior the AC and DC components separately and adding the results .
The Iact that a DC component combines with an AC component to increase the RMS is one reason
why halI-wave rectiIication should be avoided when building DC power-supplies. A halI-wave
rectiIied supply not only needs more smoothing than a Iull-wave supply, it also adds a DC
component to the current drawn Irom the mains transIormer. This increases the dissipation in the
loss resistance oI the transIormer winding, and increases the core loss by introducing a static
component into the magnetic Iield.
Some years ago, there was a period when certain power-hungry items oI consumer electronic
equipment (large colour-TV receivers in particular) were a deIinite Iire-hazard. This problem was
associated with halI-wave rectiIication, but not in an obvious way. It is oIten said that the more
pernicious the error, the more slavishly it is copied, and this is a case in point. The power-supplies
in question used bi-phase (i.e., two-diode, centre-tapped winding) Iull-wave rectiIication, which is
Iine; but someone had come up with the idea oI Iitting a Iuse in series with each diode to protect
against diode short-circuit Iailure. The problem is, that iI one oI the Iuses blows, halI-wave
rectiIication occurs, and the equipment carries on working while the mains-transIormer overheats.
The correct way to design a bi-phase power-supply is to arrange things so that the primary-side Iuse
blows iI a diode is connected across the transIormer secondary.
5
The RMS of an arbitrary waveform
An arbitrary waveIorm can always be written as a Fourier series. This has the Iorm:
V V
dc

_
n1
| V
c,n
Cos(n) V
s,n
Sin(n) |
Here we use the series to represent a voltage, but it can represent a current just as well. II we want
to Iind the RMS oI such a series we have to square it. This operation will generate a shower oI
cross-terms, but they must somehow vanish in the subsequent integration. Firstly, we have already
seen that the cross-terms produced by multiplying the DC component with a sinusoid are oI Iirst-
order, and so they average to nothing. The DC component does not interact with any oI the AC
components. Secondly, consider the Iollowing well-known trigonometric identities:
Sin(a) Cos(b) ()| Sin(ab) Sin(a-b) |
Cos(a) Sin(b) ()| Sin(ab) - Sin(a-b) |
Cos(a) Cos(b) ()| Cos(ab) Cos(a-b) |
Sin(a) Sin(b) (-)| Cos(ab) - Cos(a-b) |
Whenever two components oI diIIerent Irequency are multiplied together, they will produce a pair
oI Iirst-order terms. ThereIore oI the cross-terms will average to zero in the integration. None oI
the components interact. Hence the square oI the RMS voltage comes out as:
V
rms
V
dc

1
2a
2a
}
0

[
[

_
n1
|V
c,n
Cos(n) V
s,n
Sin(n)|

[
[

d
We obtained the substitutions needed to solve this integral earlier. They are:
Cos | 1 Cos(2) | / 2 and Sin | 1 - Cos(2) | / 2
When we make these substitutions, the Iirst-order cosine terms will, oI course, average to zero.
Hence:
V
rms
V
dc

1
2a
2a
}
0

[
[

_
n1
|()V
c,n
()V
s,n
|

[
[

d
Each remaining term in the summation, upon evaluating the integral, will come out multiplied by
2a, which will cancel with the 1/2a beIore the integral symbol. Thus:
V
rms
V
dc

_
n1
|()V
c,n
()V
s,n
|
The voltage coeIIicients aIter the summation symbol are, oI course, peak values. II we replace each
one with its RMS, then we lose the Iactors oI . Hence:
V
rms
V
dc

_
n1
|V
c(rms),n
V
s(rms),n
|
In general, iI a voltage (or current) waveIorm is made up oI an arbitrary combination oI components
6
oI diIIerent Irequency, each having an RMS value V
n
(or I
n
) when considered on its own, the true
RMS oI the combination is given by:
V
rms

\

V
dc

_
n1
V
n

When a voltage is connected across (or a current is passed through) a resistance, the power in the
resistance is given by the sum oI the powers due to the individual Irequency components.
The RMS of a floating square wave
We don't always have to know the Fourier coeIIicients in order to Iind the RMS. Notwithstanding
the Iact that we can use an electronic integrator circuit to Iind the mean oI the squares (that's how a
true-RMS meter works), there are some waveIorms which can easily
be integrated piecewise. The most obvious example is the square
wave, which is also an important special case.
Represented on the right is a voltage square-wave oI arbitrary
mark-space ratio which is AC-coupled to a resistive load. The mark
interval as a proportion oI the whole cycle is deIined as m; so that over
the course oI one cycle, the wave is positive Ior 2am radians, and
negative Ior 2a(1-m) radians. The peak positive and negative voltages
are V
1
and V
2
respectively, and since V
2
is by deIinition negative, the
peak-to-peak voltage is deIined as:
V
pp
V
1
- V
2
Because the voltage contains no DC component, its average is zero. Hence we can Iind the values
oI V
1
and V
2
relative to V
pp
by taking the integral oI the voltage between 0 and 2a and setting its
value to zero, i.e.:
2am
}
0
V1 d
2a
}
2am
V2 d 0
Thus:
| 2amV
1
- 0 | | 2aV
2
- 2amV
2
| 0
i.e.:
m ( V
1
- V
2
) -V
2
. . . . (3)
but
V
pp
V
1
- V
2
thereIore
V
2
-m V
pp
. . . . . . . . . (4)
also
V
2
V
1
- V
pp
which can be substituted into (4) to give:
V
1
(1-m) V
pp
. . . . . . . (5)
The RMS value oI the waveIorm can be obtained by piecewise integration oI the square oI the
instantaneous voltage:
7
V
rms

1
2a
2am
}
0
V
1
d
1
2a
2a
}
2am
V
2
d
Which gives:
V
rms
|m V
1
- 0| |V
2
- m V
2
|
i.e.:
V
rms
m (V
1
- V
2
) V
2
m (V
1
V
2
) (V
1
- V
2
) V
2

Using (3) as a substitution gives:

V
rms
-V
2
(V
1
V
2
) V
2

i.e.:
V
rms
-V
2
V
1
This is one simple Iorm Ior the RMS oI a Iloating square-wave which does not require prior
knowldge oI m (note that V
rms
is positive because one oI the voltages on the right-hand side has to
be deIined as negative). We obtain a more inIormative deIinition however by using equations (4)
and (5) as substitutions:
V
rms
m V
pp
(1-m) V
pp
i.e.:
V
rms
m(1-m) V
pp

The Iactor m(1-m) has the property that it is zero when m 0 or 1 (inIinitely narrow pulses contain
no energy) an is at its peak when m . This tells us that, Ior a given peak-to-peak voltage,
assuming that the waveIorm is Iloating (i.e., contains no DC component), the RMS value oI a
square-wave is at a maximum when the mark-space ratio is unity, i.e., when m/(1-m)1.
When a square wave is symmetric, V
1
-V
2
, and so the peak voltage is halI the peak-to-peak
voltage, i.e.:
V
pp
2 V
pk
Hence, Ior symmetric square waves only, i.e., when m(1-m)/ :
V
rms
(2V
pk
) / 4
i.e.:
V
rms
V
pk
The RMS oI a Iloating symmetric square wave is \2 times greater than that oI a sine wave.
The spectrum oI a symmetric square-wave contains only the odd harmonics oI the Iundamental
repetition Irequency. This is by no means the richest possible spectrum in terms oI harmonics
present, but it gives rise to the AC waveIorm which conveys the maximum amount oI energy Ior a
given peak magnitude, i.e., it achieves the limiting condition V
rms
/V
pk
1. The RMS can be
increased still Iurther however, by adding a DC component.
The RMS of a zero-based square-wave
Given the ubiquitous nature oI logic circuits, the most commonly encountered type oI square-wave
is the zero-based or clamped variety. The eIIect oI clamping one extreme oI the voltage excursion to
zero is to introduce a DC component equal to the extent to which the wave would have crossed over
the zero line had it not been clamped. Thus the DC component is -V
2
Irom the previous section,
which is given by equation (4) as:
-V
2
m V
pp
Hence the RMS oI a zero-based square wave is given by:
V
rms
m(1-m) V
pp
m V
pp
V
pp
| m - m m |
8
i.e.:
V
rms
m V
pp

V
rms
V
pp
\m
We can also obtain exactly the same result by integration over the part
oI the cycle corresponding to the 'on' or 'mark' period:
V
rms

1
2a
2am
}
0
V
pp
d m V
pp

Notice that when the mark and space intervals are equal (m), the
zero-based waveIorm has twice the RMS oI the Iloating waveIorm.
The maximum RMS Ior the zero-based waveIorm however occurs when m1; but this just means
that maximum power is delivered to a resistive load when the generator is producing a constant
output oI V
pp
, i.e., when the generator is stuck in the 'on' position.
The RMS of a noise spectrum
In the previous derivations, it has been convenient to assume that there is a harmonic relationship
between all oI the AC components oI a waveIorm. This ensures that , iI we integrate over the
Iundamental repetition interval, all oI the components to be averaged will be sampled Ior an integer
number oI cycles. This is the analytical (i.e., mathematical) RMS; but it is also possible to
determine a statistical or experimental RMS, which tends towards the true RMS when the
measurement is repeated several times and the results are averaged.
For the statistical RMS, there is no need to synchronise the sampling interval to the period oI the
waveIorm. It is suIIicient to sample the square oI the voltage or current Ior a time which is long
enough to to capture a Iew cycles oI the lowest alternating component oI signiIicant magnitude. II
the sampling interval is randomly related to a sinusoidal component, there will be a maximum, error
oI + oI a cycle. Hence sometimes the result will come out a little too high, sometimes it will come
out a little too low, but the average oI many measurements will tend towards the true value.
Furthermore, the error incurred by the lack oI synchronisation will become smaller as the number oI
cycles captured in the sampling window increases; which means that accurate measurements can be
made by sampling over a relatively long period.
It Iollows, that by a process oI electronic integration, we can determine the RMS oI a voltage or
current consisting oI a set oI harmonically-unrelated Irequency components; i.e., it is possible to
deIine and measure the RMS oI pure noise.