The RF/IF Section

The application of a linear PA in Node B design should be considered. The downlink
signal has QPSK modulation and so has greater amplitude variation than the uplink
HPSK. Accordingly, greater linearity is required. There is also the issue of a wideband
(multichannel) versus a narrowband (single-channel) approach.

In the Node B we have the option of using 12 separate RF transmitters for the 12
channels and combining their outputs at high power prior to the antenna feed, or, we
can create a multichannel signal at baseband (or IF) and pass the composite signal
through one high dynamic range, high linearity, high power amplifier. There is a large
amount of information, analysis, discussion, and speculation of the benefits of one
approach or the other, so we can confine ourselves to a review of linearizing options.
Envelope Elimination and Restoration (EER) and Cartesian approaches were
introduced in Chapter 3, so we will consider briefly other alternatives.
Linearization methods fall broadly into two categories:
 Those that use a feedback correction loop operating at the modulation rate.
 Those that use a feedback correction process to update a feed-forward correction
process, operating at a slower than modulation rate.
The former method is not particularly well suited to the wide modulation bandwidth
of 3G (5 MHz for a single channel and up to 60 MHz for a multicarrier Node B).
Whilst it is relatively simple to extract the envelope from the input RF signal and
limit the signal to give a constant envelope drive, there is some advantage to implementing
this polar split at the signal generation stage within the DSP.
In particular, it is highly likely that the transfer function through the envelope
amplifier and bias/supply modulation process will be nonlinear, and that the drive
envelope function will need to be predistorted to compensate for this error. The predistortion
factors can be held in a lookup table within the DSP, for example, and
updated if necessary by some slow feedback loop from the transmitter output.
The EER approach can yield modest improvements in linearity for quite good efficiency
(provided the envelope modulation amplifier efficiency is good). The modulation
envelope bandwidth for W-CDMA is, however, quite large (approximately 5 × 5
MHz to include key harmonics), and the efficiency of switched mode modulation
amplifiers falls off quite quickly at high switching rates.
Another method to be considered is RF synthesis. A synthesis engine converts the I
and Q (Cartesian) representation of the modulation waveform into two frequency and
phase modulated components, the vector sum of which is identical to the source signal
(see Figure 11.9). Amplification of these two constant envelope waveforms is performed
using Class C or Class F/S switching amplifiers for maximum efficiency, and
the outputs are combined to give the composite high-power RF synthesized waveform.
One of the key challenges with RF synthesis is the combination of these two
high-power FM signals without losing much of the power in the combiner process.
The vector diagram in Figure 11.10 shows how the output signal is synthesized from
the summation of the two constant envelope rotating vectors. Full output power
occurs when the two vectors are in phase. Minimum output power is synthesized
when the two vectors are 180 degrees out of phase. Using a DSP to generate the two
constant envelope phase modulated components is quite feasible, since the algorithm
is simple. The processing rate, however, must be very high to accommodate the bandwidth
expansion of the nonlinear function involved, and the sample rate of the ADCs
must also accommodate the bandwidth expansion of the FM modulated outputs.
These high sampling rates and the corresponding high power consumption of the DSP
and ADC components means that this approach is only feasible for nonportable applications
at the present time.

Alarge number of amplifier linearization solutions are based on predistortion of the
signal driving the amplifier in an attempt to match the nonlinear transfer characteristic
of the amplifier with an inverse characteristic in the predistortion process. The challenge
with predistortion is to be able to realize a predistortion element that is a good
match to the inverse of the amplifier distortion—that is, low cost and low power in its
implementation—and that can, if necessary, be adapted to track changes in the amplifier
response with time, temperature, voltage, device, operating frequency, operating
power point, and Voltage Standing Wave Ratio (VSWR).
For complex envelope modulation formats such as multicarrier W-CDMA, the envelope
excursions of the composite waveform will cause the amplifier to operate over its
full output range. This means that a predistorter element must also match this characteristic
over a wide range of input levels if high levels of linearity are to be achieved.
With a typical superhet design of transmitter, there are three locations where predistortion
can be implemented. The options are shown in Figure 11.11. An RF solution
is attractive, since it is likely to be small and does not require modification of the
remainder of the transmit stages. An IF solution is likely to make fabrication of an
adaptive predistortion element more practical. A baseband DSP based solution will
give ultimate flexibility in implementation, but is likely to take a significant amount of
processor cycles and hence consume most power.
One of the simplest RF predistorters to implement is a third-order predistorter. Recognizing
that much of the distortion in an amplifier is generated by third-order nonlinear
effects, a circuit that creates third-order distortion—for example, a pair of
multipliers—can be used to generate this type of distortion but in antiphase. When
summed with the drive to the amplifier, significant reduction in the third-order products
from the amplifier output can be achieved. Of course, good performance relies on
close matching to the gain and phase of the third-order distortion for a particular
device, and without some form of feedback control of these parameters, only limited
correction is possible over a spread of devices and operating conditions.

It would be very simple to construct an open-loop DSP-based predistorter using a
lookup table; however, for most applications, the characteristics of the transmitter
device changes so much with operating point that some form of updating of the predistortion
function is needed. As soon as an adaptive control process is introduced,
ADC components are needed, additional DSP processing used, reliable and rapid convergence
control algorithms must be identified, and the whole process becomes quite
complicated. Within a DSP, it is possible to create any predistortion characteristic
required and rapidly update the transfer function to follow changes in the amplifier
device response.
As the cost and power consumption of DSP engines continues to fall and the processing
power increases, the digital baseband predistortion solution becomes more
and more attractive—first for Node B use but also for portable use. There are two main
options for updating a predistortion lookup table: using power indexing, which
involves a one-dimensional lookup table, or Cartesian (I/Q) indexing, giving rise to a
two-dimensional lookup table.
Power indexing will result in a smaller overall table size and faster adaptation time,
since the number of elements to update is smaller. It does not, however, correct AMPM
distortion, which means that only limited linearization is possible. I/Q indexing
will provide correction for both AM-AM and AM-PM distortion and so give optimum
results, but the tables are large and adaptation time slow. For wideband multicarrier
signals it is necessary for the lookup table to have a frequency-dependant element to
accommodate frequency-dependent distortion through the amplifier chain. This can
give rise to three-dimensional tables.
In summary, baseband digital predistortion is the most versatile form of predistortion
and will become more widely used as the cost and power consumption of DSP
falls. Because of the slow adaptation time for a lookup table predistorter, it is not
possible to correct for the memory effect in high-power amplifiers, and so this will
limit the gain for multicarrier wideband applications. Correct choice of lookup table
indexing will give faster adaptation rates and smaller table size; however, frequency
dependent effects in the amplifier cannot be ignored.
An alternative to using a lookup table is to synthesize in real time the predistorter
function, much like the third-, fifth-, and seventh-order elements suggested for RF predistortion.
This shifts the emphasis from lookup table size to processor cycles, which
may be advantageous in some cases.
The final linearization method to be considered is the RF feed-forward correction system.
This technique is used widely for the current generation of highly linear multicarrier
amplifiers designs in use today, and there are many algorithm devices for correcting
the parameters in the feed-forward control loops. More recently, combinations of feed
forward and predistortion have appeared in an attempt to increase amplifier efficiency
by shifting more of the emphasis on pre-correction rather than post-correction of distortion.
Afeed-forward amplifier operates by subtracting a low-level undistorted version of
the input signal from the output of the main power amplifier (top path) to yield an
error signal that predominantly consists of the distortion elements generated within
the amplifier. This distortion signal itself is amplified and then added in antiphase to
the main amplifier output in an attempt to cancel out the distortion components.
Very careful alignment of the gain and phase of the signals within a feed-forward linearization
system is needed to ensure correct cancellation of the key signals at the input
to the error amplifier and the final output of the main amplifier. This alignment involves
both pure delay elements to offset delays through the active components, as well as
independently controlled gain and phase blocks. The delay elements in particular must
be carefully designed, since they introduce loss in the main amplifier path, which
directly affects the efficiency of the solution. Adaptation of the gain and phase elements
requires a real-time measurement of the amplifier distortion and suitable processing to
generate the correct weighting signals. Most feed-forward amplifiers now use DSP for
this task. Where very high levels of linearity are needed, it is possible to add further control
loops around the main amplifier. Each subsequent control loop attempts to correct
for the residual distortion from the previous control loop, with the result that very high
levels of linearity are possible but at the expense of power-added efficiency through the
amplifier. Feed-forward control requirements are shown in Figure 11.12.
In summary, feed-forward amplifiers can deliver very high levels of linearity over
wide operating bandwidth and can operate as RF-in, RF-out devices, making them
attractive standalone solutions. Their main drawback is the relatively poor efficiency.
Many new multicarrier amplifier solutions are utilizing predistortion correction techniques
to try and reduce the load on the feed-forward correction process so that it can
operate in a single-loop mode with good main amplifier efficiency. The poor efficiency
makes feed forward an unlikely candidate for handset applications; however, since
these tend to operate only in single-carrier mode, predistortion techniques alone are
likely to give sufficient gain. 255