You are on page 1of 8

Self-Cancellation of Third-Order Modulation Products

By Cecil Deisch, Tellabs Operations, Senior Staff Engineer, Naperville, Ill.



Abstract
It seems counter-intuitive, but the second-order nonlinearity of an amplifier can be
advantageously used to greatly reduce its troublesome third-order sidebands. Additionally, there
often exists a set of circumstances which allow a simple short circuit to automatically cancel a
substantial amount of these unwanted third-order sidebands. Thus, for this situation, there is no
need for complicated pre-distortion schemes. Much of this is already well-known, but heres a
simple, basic explanation.

Introduction
Because of nonlinearities, all amplifiers generate intermodulation products, some of which are
very difficult to remove. Take the case of a narrow-bandwidth amplifier which must put out a
very clean RF signal; that is, any frequencies outside the band due to intermodulation must be
greatly attenuated, often by 50 dB or more. For modulation products far away from the desired
RF band, its often reasonable to filter them out. The well-known modulation products caused
by second-order nonlinearities generate unwanted frequencies at DC, at baseband, and at the
second harmonics of the in-band signals. All these can be effectively filtered out because of
their great distance from the in-band signals.

However, some intermodulation frequencies caused by third-order amplifier nonlinearities are
especially difficult to get rid of. These troublesome frequencies are of the form of (2F
1
F
2
)
where F
1
and F
2
are any two frequencies of the in-band RF signals. Since the RF band is narrow,
F
1
& F
2
are quite close together, so their intermodulation frequencies also lie very close to the RF
band (and even within the RF band itself).

Since it is not possible to filter these unwanted close-in frequencies, if the signal level of the
RF amplifier is high enough where the third-order modulation products become troublesome, the
usual remedy is to back off the RF levels enough to meet stringent linearity requirements. This
is a very costly remedy, because higher power amplifiers must be used (at extra cost and extra
heat generation) to get the required RF output power.

As a result of this, much effort has been put forth to improve the amplifier linearity by pre-
distorting the input and/or post-distorting the output of the amplifier in order to cancel some
of the amplifier nonlinearities and get a bit more output power with the required linearity. Some
of these schemes are quite complicated and yet bring only limited improvements.

Some Basic Facts
Most linear single-ended narrow-band RF amplifiers are operated with just a small amount
of nonlinearities. That is, at very low signal levels, the amplifier is essentially linear and there
are insignificant intermodulation products. As the signal level is increased, the first
nonlinearities to become apparent are the second-order products. That is, second harmonics and
baseband frequencies appear. Further signal level increases make the third-order nonlinearities
become noticeable, including the especially troublesome close-in third-order frequency products
2
directly adjacent to the RF band. Even higher signal levels cause fourth- and higher-order
nonlinearities to appear, however, operation at these high signal levels is rarely allowed because
the lower-order nonlinearities are already becoming troublesome.

Example:
Although it doesnt seem to immediately apply to a three-terminal amplifier, consider first a
simple two-terminal nonlinear device which has narrow-
band AC current applied to it and the resultant AC
voltage has both 2
nd
& 3
rd
order nonlinearities.

For simplicity, suppose just two frequenciesat the
upper and lower edges of the bandrepresent the in-
band signal: a sine signal of amplitude 1 and frequency
F
1
=100 Hz and another sine signal of amplitude 1 and
frequency F
2
=110 Hz. Assume the nonlinear device
has a normalized (time- and frequency-independent)
transfer function of y = x + Ax
2
+ Bx
3
where y is the output voltage and x is the input current
consisting of the sum amplitude of the two frequencies F
1
& F
2
. For the following facts, Ive
chosen to eliminate the mathematical clutter because its just a huge amount of routine algebraic
manipulation and substitutions of the prosaic sine and cosine trig functions.
200 220
9
0
1
0
0
1
1
0
1
2
0
10
DC
Baseband
(2
nd
order )
Close -in
(3
rd
order )
2
nd
Harmonic
(2
nd
order )
Desired
Narrow -Band
Close -in
(3
rd
order )
Device Output with Nonlinearities
frequency

Fact 1: The first term of the transfer function represents linear operation and it generates no
intermodulation products. The output contains only those frequencies present on the input.

Fact 2: The second term represents the amplifier second-order nonlinearity and it causes the
output to have a DC component (F
1
F
1
) and (F
2
F
2
), a difference frequency of 10 Hz
(F
2
F
1
) and (F
1
F
2
), and higher frequencies of 200 Hz (F
1
+F
1
), 210 Hz (F
1
+F
2
), and 220 Hz
(F
2
+F
2
). All these frequencies are far from the narrow band comprised of F
1
& F
2
and can be
reasonably filtered out. Note, in particular, that this second term generates no frequencies close-
in to the desired RF band.

Nonlinear
Device
Input
Current
Output
Voltage

3
Fact 3: The third term of the transfer function represents third-order nonlinearity and causes the
output to have many intermodulation products, for example: 300 Hz (F
1
+F
1
+F
1
), 310 Hz
(F
1
+F
1
+F
2
), etc., all of which can be reasonably filtered out. However, the output also
contains troublesome close-in products of 90 Hz (F
1
+F
1
F
2
) and 120 Hz (F
2
+F
2
F
1
). These
modulation products are so close to the desired RF band that they cannot be reasonably filtered
out. Note that all other third-order intermodulation products are far away from the RF band and
that, in particular, there are no products at DC or baseband.

Fact 4: It is a happy coincidence that there exists a single 10 Hz sinusoid with appropriate
amplitude and phase which, when applied to the input of the transfer function above, exactly
cancels the close-in frequencies of 90 and 120 Hz. This has long been known and forms the
basis of some harmonic canceling schemes. When this 10 Hz sine is applied, it is the second-
order nonlinearity of the transfer function which cancels the third-order close-in frequency
components. That is, the second-order nonlinearity creates 90 Hz =(F
1
10 Hz) and 120 Hz =
(F
2
+10 Hz) which cancel the troublesome third-order close-in amplifier nonlinearities.

Fact 5: If there is a baseband short circuit presented at the device output, there is a specific set of
coefficients for the nonlinearity which will automatically cancel the close-in third-order
frequency components. That relationship is:
3 2
) 866 (. Bx x B x y + = where B is positive.
This means that if this baseband short-circuit current is fed back to the input, the close-in third-
order components will be exactly canceled. Another nice feature is that if the signal amplitude
changes, which of course causes the amplitude of all the intermodulation products to change, the
short-circuit current automatically adjusts such that the third-order close-in frequencies are still
canceled.

Fact 6: It is another happy coincidence that the nonlinearities of real-world components often
closely approximate the ideal coefficient relationship in #5 which means that by applying a
baseband short circuit to the amplifier output; the troublesome close-in third-order components
are greatly reduced. This simple cancellation method requires no special signal pre-conditioning,
is self-adapting and is automatic.

Fact 7: It is also well-known that by shorting the second-harmonic frequencies, that the close-in
third-order frequency components can be attenuated, although usually not by nearly as much as
is possible by shorting the baseband.

Fact 8: Notice that for balanced or push-pull amplifiers, although they to generate close-in third-
order modulation products, the technique of applying a baseband short wont work for canceling
these close-in frequency components for the simple reason that balanced amplifiers essentially
have no second-order nonlinearities. Thus, there is no way to inject a baseband signal and have
it modulated up to cancel the close-in frequencies. In fact, one of the celebrated benefits of a
balanced amplifier is that second-order nonlinearities tend to cancel (however, the third-order
nonlinearities add).

A Simple Experiment
Using the same two input signals F
1
& F
2
as
above, and because of Fact 4, a simple experiment of
Nonlinear
Device
Input
Current
Output
Voltage
10 Hz
Injection
Current

4
injecting a 10 Hz baseband signal is very revealing. Starting with a small-amplitude 10 Hz
injection signal and varying its phase, it is found that the phase which decreases the resultant 10
Hz baseband signal also reduces the troublesome close-in 3
rd
-order components. While keeping
this optimum phase, as the amplitude of the 10 Hz injection signal is increased, there comes a
point where the close-in 3
rd
-order components are essentially completely cancelled. Depending
on the magnitude of the 2
nd
-order coefficient A, at this optimum injection amplitude, the
resulting 10 Hz 2
nd
-order component may still be present, albeit at a lower amplitude than with
no baseband injection 10 Hz signal.

The other interesting result of this experiment is that the phase of the original 2
nd
-order
baseband output voltage distortion component is such that if a resistor is placed on the output of
the device, the current through the resistor has exactly the phase needed to reduce the 2
nd
-order
as well as the close-in 3
rd
-order components. (Of course, this is a frequency-limited resistor; it
looks like a pure resistance at baseband, but is open-circuit at higher frequencies.) As this
resistor value is reduced, distortion components are also reduced. And, if the distortion
coefficients satisfy the relationship of Fact 5, when this resistor value reaches zero (a short), both
the close-in 3
rd
-order and the baseband 2
nd
-order components are completely cancelled.

Nonlinearities of the Simple Diode
Suppose a simple diode is driven by an AC signal current (e.g., the F
1
& F
2
above), what is the
resultant voltage, including its nonlinearities? Although this is certainly not a practical amplifier,
it shows that a real-world componentwith strong nonlinearitieshas a very attractive transfer
function; one that is amenable to harmonic cancellation. An ideal diode has an equation of the
form:
M
V
e I =
Therefore,
( ) I M V ln =

Next, create the series expansion about some DC bias current I
0

+ + + =
5
0
5
4
0
4
3
0
3
2
0
2
0
0
5 4 3 2
) ln(
I
x
I
x
I
x
I
x
I
x
I M V
Where x represents the AC or incremental current about DC bias of
0
I
Normalize by deleting the DC term and multiplying by
M
I
0

+ + =
4
0
5
3
0
4
2
0
3
0
2
5 4 3 2 I
x
I
x
I
x
I
x
x y
Where y represents the AC voltage caused by AC current x.

First, consider the transfer function which contains only the 2
nd
and 3
rd
order nonlinearities:

2
0
3
0
2
3 2 I
x
I
x
x y + =
5

Surprisingly, the coefficients of the 2
nd
and 3
rd
nonlinearities for this ideal diode exactly satisfy
the special coefficient relationship (see Fact 5):

0
2
0
2
1
3
1
866 .
I I
=

This special coefficient relationship of the transfer function means that a short-circuit at the base-
band frequencies will exactly cancel the close-in third-order intermodulation products. This
implies that, no matter how large the magnitude of the output signal, there will be no close-in
third-order intermodulation products. In reality, the signal magnitude cannot be raised without
limit because eventually the fifth-order and other intermodulation products will create unwanted
close-in products, and further, the series expansion above does not converge for x>I
0
. Of
course these are idealized results, but they suggest tantalizing prospects.

Different Diode Results
The very good transfer function obtained above is for the voltage across a current-fed diode;
however the transfer function for the current through a voltage-fed diode is quite different and
very enlightening. For the transfer function of the current-fed diode above, y represented the
voltage and x represented the current fed into the diode. Its easy to recast the transfer function
for a voltage-fed diode by performing a reversion of the series. The result is:

2
3 2
6 2
o o
I
y
I
y
y x + + =
Here x represents the current through the diode when a voltage y is placed across the diode.
(This might represent the situation of driving the base-emitter junction of a bipolar transistor.)
First, note that the 2
nd
- and 3
rd
-order coefficients no longer satisfy the ideal relationship of Fact
5 because the 2
nd
-order coefficient is about 2 too large. Since this is a voltage-fed nonlinear
device, the normal baseband shunt short circuit must be replaced by a baseband open circuit in
series with this diode. As a consequence, one finds that placing a baseband open in series with
this diode does not cancel the close-in 3
rd
-order current products; in fact, they appear to be
unchanged. The reason is that by placing a series open circuit at the baseband, there is so much
over-compensation that the close-in 3
rd
-order current products have reversed phase, but remain at
the same amplitude.

All is not lost however and its still possible to get self-cancellation of the 3
rd
-order products
by using a baseband resistor instead of an open. In fact, for this specific normalized transfer
function, placing a 1 resistor (at the baseband frequency; a short at all higher frequencies) in
series with the diode gives near-perfect close-in 3
rd
-order cancellation while leaving the 2
nd
-order
baseband current component at about half its original amplitude. Its now obvious why the
original open gave too much compensationtwice as much in factwhereas the resistor gives
just the right amount.

Calculating the Optimum Value Baseband Shorting Resistance
6
As shown in the examples above, a baseband shorting resistor of zero ohms may give near-
perfect cancellation of the close-in 3
rd
-order modulation products (the voltage across an ideal
current-fed diode) or may appear to give no benefit at all (the current through an ideal voltage-
fed diode).

There is a simple way to calculate the optimum value of the baseband shorting resistor just by
knowing the nonlinearity coefficients of the normalized transfer function. The nonlinear transfer
function of a candidate RF amplifier can be derived in several ways based on measured data.
But first observe that an amplifier with known coefficients A and B (of a normalized transfer
function), produces close-in 3
rd
-order sidebands of amplitude of .75B and a 2
nd
-order baseband
amplitude of A. So perhaps the easiest way to obtain the coefficients of a candidate amplifier is
to apply two tones of equal amplitude whose frequencies lie at the upper and lower edges of the
RF band. The amplitude of the two tones is increased until the amplitude of the close-in 3
rd
-
order sidebands are high enough to reliably measure. For example, if the 3
rd
-order sidebands are
40 dB down from the in-band tones (that is, their normalized amplitudes are .01 volt), then the
value of B is .0133. At this same setting, suppose the baseband amplitude is measured at 18 dB
down from the in-band tones (that is, its normalized amplitude is .126 volts), the value of A is
also .126. Thus, the normalized transfer function of this candidate amplifier is

y = x .126x
2
+ .0133x
3

Once the normalized values of the nonlinear coefficients are known, the optimum value of the
baseband resistance is:

1
75 .
2
=
B
A
R


For the example amplifier above, the optimum value of baseband resistance is
=

= 59 . 1
) 0133 (. ) 75 (.
126 .
2
R

Note that the formula for the ideal baseband resistance allows for negative resistance,
depending on the values of the nonlinear coefficients.

The FET Amplifier
Of course, a FET amplifier is a three-terminal device quite unlike the two-terminal devices
used as examples above. However, it turns out that the nonlinearities of a FET amplifier are
mostly caused by variations in transconductance due to changes in output voltage (voltage from
drain to source) and by changes in output current (drain current). Experience indicates that
nonlinearities caused by the input are usually much smaller. Because of this, its nonlinear
performance therefore closely resembles a two-terminal device and most of the benefits of output
nonlinearity reduction are realized by placing a baseband short (or resistance) on the output.

Conclusions
7
The examples above have used only two frequencies, F
1
& F
2,
to demonstrate the inter-
relationship between the 2
nd
-order and 3
rd
-order distortion products because, even with this
simplification, the math is quite tedious. However, through the use of SPICE simulations, the
conclusions above apply equally well to any number of discrete frequency components in the
narrow-band RF signal. By understanding the fundamental nature of this self-cancellation
scheme, other potential prospects suggest themselves:

1. The 2
nd
- & 3
rd
-order coefficients of the real-world transfer function may be such that a
direct baseband short on the output of an amplifier is not optimum. That is, perhaps a low-
value (non-zero) baseband resistance would be better.
2. On the other hand, a low-value but negative baseband resistance might give better harmonic
cancellation results.
3. Since different biasing points give different nonlinearity coefficients, the most optimum
bias point for minimum distortion without baseband correction may not be the best bias
when using baseband correction. That is, when using baseband correction, there may be a
slightly better bias point to give the lowest overall distortion.
4. If a single-stage amplifier has a nearly 180-degree phase shift from input to output, then
perhaps an appropriate resistive baseband feedback from output to input might be able to
cancel most of the distortions caused by both input and output nonlinearities. Of course, if
the phase shift is significantly different than 180 degrees, the feedback impedance and
phase would have to compensate.
5. The examples (above) of nonlinearities in the ideal diode might be used for handling the
nonlinearities of the base-emitter junction of a bipolar transistor amplifier.

Keep in mind that this baseband cancellation scheme works best for gentle nonlinearities
(the diode examples above are gentle); that is, there are no abrupt changes in the transfer
function such as saturation or cutoff. Abrupt changes quickly bring about much larger 5
th
- and
higher-order odd harmonic nonlinearities which not only generate additional close-in sidebands,
but they also cannot be cancelled by a baseband short. This just reflects practical experience that
once an amplifier reaches abrupt limits, you cannot get any more output power. But, even if this
amplifier is operating near some abrupt limit, the 3
rd
-order sidebands can still be attenuated with
baseband shorting. However, beyond a certain amount of 3
rd
-order attenuation, the 5
th
& 7
th

close-in intermodulation products will then be the dominant sidebands.

This analysis also does not include the effects of small phase shifts which may be present in
the transfer function of real amplifiers. This implies that a bit more improvement in linearity
might be gained by introducing a small amount of reactive impedance in the baseband shorting
resistance. Despite this possible phase shift, this analysis gives the bulk of the improvement.

About the Author

Cecil W. Deisch received his MSEE degree from New York University and has spent most of his
career at AT&T Bell Labs, concentrating mostly on magnetics and analog circuit design. In
addition to being awarded several patents, he has published many technical papers in various
branches of electrical engineering and is an AT&T fellow. During his career, he has designed
8
circuits covering the frequency range from 20 Hz (telephone ringing voltage) to 1 GHz. He now
works for Tellabs, where he developed this new low loss coupler configuration.