Technical Article, Click to open in new window

New RF Metrics for the
Smartphone-centered
World
A
decade ago, “cell phones” were all about
voice communication, an extension of
the home phone for those who needed
to be on the go. Global roaming was minimal
at best, with multiple communications standards worldwide: GSM, TDMA and CDMA,
primarily, with many devices carrying backward compatibility to AMPS. To match these
applications, RF power amplifier (PA) performance was all about peak power capability.
Performance at 3.4 V was the key metric as it
drove output power capability and, as a result,
a handful of parameters based on modulations
characteristics were evaluated along with the
key product differentiator, power added efficiency (PAE). Essentially, suppliers and customers alike utilized “static” measurements
and kept evaluations very straightforward.
Fast-forward to today’s Smartphone-centered world and one will quickly find that these
“static” measurements are not nearly as relevant; they are evaluating more of the exception than the rule. The implication is that these
simple metrics are driving incorrect tradeoffs
in cellular front-end and PA design. Today’s
Smartphone-centered world deserves Smartphone-centered front-ends. These front-ends
operate in a dynamic world and need new, “dynamic,” Smartphone-centered metrics to succeed.
Smartphones = Mobile Data
First and foremost, while a Smartphone can
simply be described as a new generation of mobile devices, the mobility attribute has expanded dramatically. Rather than simply being an
extension of voice communication when someone is away from their home or desk phone,
mobile phones are quickly becoming the only
phone owned by many individuals. And, with
the growing coverage of GSM and W-CDMA
networks, mobile now also means global roaming—across multiple countries and multiple
mobile operators. Secondarily, while still a tool
for voice communications, Smartphones are
predominately data-driven devices—whether
users are surfing the Internet, downloading the
latest video or song, or twittering about their
latest favorite restaurant.
The RF implications of these new usage
models are quite dramatic. Smartphone users seek the longest operating time between
charges, given their heavy data usage during the day, so making the most of a battery’s
full range of operation is increasingly important. Next, the form factor of these devices
has decreased dramatically despite increased
semiconductor content. The net result is an
Ben Thomas and Jackie Johnson
RFMD, Greensboro, NC
Reprinted with permission of MICROWAVE JOURNAL® from the January 2011 issue.
©2011 Horizon House Publications, Inc.
Special Report
It Is All about Battery Current Draw These
Days
In the past, efficiency was used as the primary metric for
a PA’s performance in a cellular application, particularly at
peak power and a single voltage, traditionally 3.4 V. In reality, however, a mobile phone’s battery does not stay at a
constant voltage; in fact, as illustrated in Figure 1, a lithiumion battery’s voltage changes based on the percent charge.
The battery voltage typically ranges from around 4.2 to 3
V, depending on the amount of charge available. When one
examines the battery discharge curve in detail, data reveals
that for over 80 percent of the operating time the battery
voltage is above 3.4 V—the first evidence that the traditional
measurement point is the exception not the rule.
Looking back to the use of efficiency as an indicator of
performance, efficiency has been the staple for comparing
RF front-ends, although they require a subjective choice
of a single voltage point. This metric is convenient to measure and evaluate, but does not represent real-world usage models. For example, the calculation for efficiency can
be used to demonstrate the delta in performance between
traditional GSM PA solutions that do not incorporate DCDC converters because these PAs utilize a fixed collector
voltage for operation.
Pout
10 10
Ef f% = 1000 ∗100
Vcc Icc
(1)
As a result, at a battery voltage of 3.4 V while delivering 31
dBm of power at the PA output, the efficiency is ∼35 percent,
which is typical for solutions today. By changing the battery
voltage to 3.8 V the calculation for efficiency shows a degradation of over 4 percent, which is a result of the energy
being dissipated thermally. While this shows a reasonable
indication of the potential thermal implications of the solution, efficiency’s dependence on battery voltage does not give
a good indication of the solution’s effect on battery life and
the resulting end user’s talk-time. Specifying RF front-ends
in terms of current consumption (mA) is the only true way to
compare solutions as to their impact on talk time since batteries are specified by their capacity in mA/hours. Therefore, a
better comparison, which more closely resembles real-world
VBATT (V)
usage models, is to
4.5
calculate average
4.0
current consump3.5
3.0
tion across the dis2.5
charge curve. Ad2.0
1.5
ditionally, this will
1.0
take into account
0.5
0
the possibility of a
100
80
60
40
20
0
solution that utiCHARGE LEVEL OF BATTERY (%)
lizes a DC-DC
converter to adjust s Fig. 1 Li Ion battery discharge curve.
PA collector voltPOUT ANT = 29 dBm
age with output
1.2
power—a
com1.0
mon theme in to0.8
day’s Smartphone
0.6
RF solutions—and
0.4
provide a way to
0.2
better compare the
0
4.2 4.0 3.8 3.6 3.4 3.2 3.0
benefits of the two
VBATT (V)
types of implementations.
s Fig. 2 Current vs. battery voltage for
In terms of solutions with no DC-DC converter.
measuring
current consumption, a PA designer could, in a straightforward fashion, measure current over the operating voltage
range. In a traditional GSM PA not connected to a DCDC converter, we find, intuitively, that current consumption remains constant over the battery voltage range (see
Figure 2).
However, in a solution that utilizes a DC-DC converter
to supply the PA collector voltage, we find that current
consumption varies with battery voltage. Thus, since the
battery voltage varies with charge level, we can measure
the current consumption relative to the battery charge
level (see Table 1).
IBATT (A)
extremely compact design that is unable to mask the heat
generated by inefficient operation; thus, thermal performance has jumped to the forefront of consideration in new
components. The use of DC-DC converters to address
these thermal concerns is also on the rise. Third, an understanding of the usage models of the global cellular communication standards indicate a much more dynamic output
power model for mobile devices, with peak power being
more of the exception than the rule. Combining these factors with global roaming across various network operators’
infrastructure deployment strategies creates a need for the
RF to be the most efficient for any given power level the
network dictates. Last but not least, Smartphones are data
devices by design. With the multitude of data modulations
driving various linearity and power requirements, dynamic
assessments of performance are mandatory to fully evaluate an RF front-end’s or a PA’s fit for the application.
TABLE I
measured data of rf front-end with DC-DC
converter
Charge
Level (%)
Vbatt (volts)
Ibatt (amps)
DC Pin
100
4.2
0.600
2.52 W
92.3
4.1
0.614
2.51 W
83.9
4.0
0.628
2.51 W
72.3
3.9
0.643
2.50 W
53.1
3.8
0.659
2.50 W
30.8
3.7
0.675
2.49 W
20.3
3.6
0.693
2.49 W
13.8
3.5
0.712
2.49 W
9.51
3.4
0.733
2.49 W
5.98
3.3
0.754
2.48 W
4.27
3.2
0.778
2.48 W
2.88
3.1
0.802
2.48 W
2.21
3.0
0.829
2.48 W
1.54
2.9
0.857
2.48 W
0
2.5
---
Special Report
Thermal Concerns Are on
the Rise
Taking the data from the above
example, similar methodologies are
used to evaluate the impact of varying current consumption on thermal performance (see Figure 5).
IBATT (A)
0.8
IBATT CURVE FIT
0.7
0.6
0.5
100
80
60
40
20
0
LI ION BATTERY CHARGE LEVEL (%)
s Fig. 3 Data graphed with a 10th order
polynomial curve fit applied.
SINGLE-ENDED NO
DC-DC CONVERTER
SINGLE-ENDED WITH
DC-DC CONVERTER
0.60
0.55
IBATT (A)
P1 11 P2 10
∫ Ibatt(L)dL = 11L + 10 L +
P3 9 P4 8 P5 7
L +
L +
L +
9
8
7
P6 6 P7 5 P8 4
L +
L +
L +
6
5
4
P9 3 P10 2
L +
L + P11L
(2)
3
2
Computing the equations reveals that this solution will give
the designer an average current of
Ibattavg=0.665A.
Comparing the two GSM PA solutions delivers a calculated, nonsubjective measurement of relative
performance, along with metrics that
are directly and easily correlated to
battery life and talk-time. In this case,
the solution that utilizes a DC-DC
converter can more efficiently utilize
current for the given, backed-off power level (ECTEL power of 29 dBm),
resulting in ~400 mA less average
current consumption. A Smartphone
RF architecture that values GSM current consumption at backed-off power
levels would quickly see a significant
advantage in implementing a DC-DC
converter.
Conversely, if we stayed with a traditional measurement of peak power
at 3.4 V, the solution with a DC-DC
converter would show worse current consumption. However, again,
this single measurement point does
not adequately depict the situation.
As shown in Figure 4 (similar to the
backed-off power example), when one
considers the average current over
the battery voltage and weight that
given the percentage of time a battery
is at the voltages, one can intuitively
see that, given normal mobile phone
operation, the better performing solution is one that utilizes a DC-DC
converter.
0.9
0.50
0.45
0.40
0.35
0.30
3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4
VBATT (V)
s Fig. 4
voltage.
Current consumption vs. battery
SINGLE-ENDED NO
DC-DC CONVERTER
SINGLE-ENDED WITH
DC-DC CONVERTER
PDIS (W)
Next we graph the current consumption over the charge level (see
Figure 3.). By taking the polynomial
and integrating the curve, it gives us
Equation 2
1.9
1.7
1.5
1.3
1.1
0.9
0.7
0.5
0.3
3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4
VBATT (V)
s Fig. 5
Power dissipated vs. battery voltage.
Based on the conservation of power
we quickly see that over the majority of the battery discharge curve of
a non-DC-DC converter-based solution will dissipate dramatically higher
thermal energy. Approximately 1.5
to 2 W of heat being dissipated in
the Smartphone is easily noticed by
the end user depending on the form
factor and thermal properties of the
Smartphone. Since W-CDMA is full
duplex (meaning the system transmits
and receives simultaneously), the PA is
powered on and transmitting for the
full duration of the data call. GSM, on
the other hand, is time-based, resulting in 1/8 duty cycle of the PA up to
4/8 for GPRS multi-slot operation. In
a contrasting case, if the W-CDMA
PA consumes 400 mA of current, this
is a constant requirement from the
battery. In a GSM system, if the PA
consumes 1.5 A of current and the
duty cycle effect is applied, this would
result in less than 200 mA of average
current. Examining these two types of
systems, it is easy to understand why
DC-DC converters have been broadly
adopted in full duplex systems—to
help minimize this thermal dissipation.
With these examples, it is clear that
measuring the thermal impact of a PA
solution—no matter the communications standard—is best done by evaluating an average thermal dissipation
over intended use range of the battery
discharge curve.
Moving People + Fixed
Infrastructure = Varying
Output Power
While this concept of output power
varying as a mobile phone moves with
its user seems intuitive, still it remains
that for the last decade the dominant
majority of metrics have been related
to peak power levels. The CDG curve
developed for CDMA-based mobile
operators was the first attempt to
more adequately depict what is really
happening with a mobile phone’s output power during typical operation.
GSM Association quickly came up
with a similar metric, as the expansion
of W-CDMA modulation showed similar attributes, and the DG.09 curve,
which estimates the probability density of varying output powers for WCDMA voice modulation, was created
(see Figure 6).
Despite the fact that both the CDG
and DG.09 curves have been around
for a while, there remains a large focus on single power level metrics,
which can quickly drive the wrong
decision points. For example, let us
compare two completely different
W-CDMA PA control architectures:
a single-ended W-CDMA PA that has
three distinct power modes (PA “A”);
and a quadrature W-CDMA PA that
utilizes continuous analog bias control and collector voltage adjustment
coming from a buck DC-DC converter (PA “B”). Often these two types
of solutions would be measured at 0
and 24 dBm (peak) output power to
see which has the best performance.
Special Report
-38
0.5
the exact set points for the creation
of PDF curves, we believe that utilizing a set of PDFs, however defined, is
far better than utilizing a few discrete
points to make key architectural and
component selection decisions.
No More Excuses for Doing
it the Right Way
Batteries do not stay at the same
voltage during the discharge cycle,
thus a dynamic environment exists in
the mobile phone. By definition, mobile phone users are mobile, providing another, simultaneously dynamic
environmental variable. Although we
have spent so many years using static
measurement points to determine the
“goodness” of a front-end solution,
one quickly sees that these assumptions are not in line with how the real
cellular-based mobile world works.
14
12
10
P(X) (%)
to increase power,
resulting in higher
Current consumption comparison of two W-CDMA
signal to noise ratio
architectures
(SNR) at the receivW-CDMA PA
0 dBm Icc
24 dBm Icc
DG.09 Icc
er, thus limiting the
Architecture
BER and error correction. This results
PA “A”- SE, 3 power9 mA
420 mA
26 mA
mode
in faster data rates.
To begin measurPA “B”- Quadrature,
12.5 mA
460 mA
22 mA
ing average current
analog bias and
collector voltage
consumption with
control
these data rate implications in mind,
Table 2 shows that the difference is
a 3G data PDF is shown in Figure 7
performance, and, at quick glance,
and Table 3 to measure the dynamic
one can see the detriment of using a
performance of a W-CDMA PA.
“static” metric. If 0 dBm or peak curLooking to the future and 4G LTE
rent consumption was used, PA “A”
adoption, it would be prudent not to
might be chosen. If the metric for
make the same mistake again by asdecision making was DG.09, which is
suming “static” measurement points
a better representation of W-CDMA
as history tells us that this is a highly
voice performance, the decision makunlikely scenario. As such, we have
er chooses PA “B,” while also gaining
evaluated the systems implications of
the benefits of a quadrature PA soluLTE’s QPSK modulation schemes and
tion—VSWR tolerance and broadsee a shift, yet again, for real-world
band capability.
implementation in mobile devices.
Table 4 outlines the evaluation points
Smartphones = Data
for this higher order data modulation
Thus far we have centered the disscheme and Figure 8 serves as a comcussion of metrics on voice-centric
posite of the three different PDFs
performance. Smartphones, however,
presented.
spend the majority of their time in
A quick analysis of Figure 8 shows
a data mode. With each new Smartthe current and upcoming complexity
phone rollout we are seeing an inof Smartphone RF, and hopefully it
crease in the number of data modes—
highlights the need to abandon “statbasic W-CDMA data, then HSPA
ic” measurements of performance.
and now HSPA+. Each mode carries
Even if there is disagreement as to
different performance requirements
and demands on the PA solutions. In
TABLE III
short, as the data rate goes up, the
implication on power output is on the
3G data PDF evaluation power
rise in the center point of the output
levels and probabilities
power probability density function
Pant (dBm)
3G.DATA [3G HSPA+]
(PDF). For example, although little
P(x) %
to no formal data has been published,
24
8.8
in speaking with many customers and
21
5.3
cellular platform providers there is
18
8
general consensus that, on average,
where DG.09 is “centered” around
15
10.6
0 dBm, higher order data modula12
12.2
tions are centered between 10 to 15
9
12.9
dB higher based on the needs to ensure sufficient data throughput. This
6
12.2
requirement is logical as the higher
3
10.6
order modulation formats place more
0
7.9
symbols in the constellation. As more
-3
5.3
symbols are packed in the constellation there is less error in phase and
-8
3.5
amplitude between the symbols re-18
1.5
sulting in higher bit error rates (BER).
-28
0.7
This requires the transmit (TX) system
TABLE II
8
6
4
2
0
–54 –42 –30 -18 -6
6
PANT (dBm)
s Fig. 6
function.
18
30
DG.09 probability density
TABLE IV
4G data PDF evaluation power
levels and probabilities
Pant (dBm)
4G.DATA [LTE QPSK]
P(x) %
24
5.3
21
3.5
18
5.3
15
8
12
10.6
9
12.2
6
12.9
3
12.2
0
10.6
-3
7.9
-6
5.3
-11
3.5
-21
1.5
-31
0.7
-41
0.5
Special Report
14
DG.09
3G.DATA (3G HSPA+)
12
14
12
8
10
6
P(X) (%)
P(X) (%)
10
4
2
0
–54 –42 –30 -18 -6
6
PANT (dBm)
s Fig. 7
function.
8
6
4
18
30
3G data probability density
There might be some who say, “But
the measurement equipment and
time to evaluate such dynamic profiles did not exist.” Perhaps not at that
time, but they certainly do now. At
RFMD®, there are component characterization systems that can evaluate
a multi-mode, multi-band environment over all temperature, load and
power level conditions in a matter of
days; the same amount of information
collection would have taken a matter
of months previously. The technology
is available to allow us to be more exacting in our analysis.
2
0
–54 –42 –30 -18 -6
6
PANT (dBm)
s Fig. 8
curves.
18
30
Composite of 3G and 4G PDF
Perhaps most important is that we
are at a huge inflection point in our industry. For an industry that has been
driven largely by voice, under a minimum number of cellular modulation
schemes, the world is changing fast.
The growth of Smartphone volume
is of the kind of segment growth we
have not seen in a decade. And with
these new devices comes an extremely dynamic, complex and demanding
operating environment. Static, single
voltage, single power level measure-
ments to determine the goodness of
a solution are archaic at best. Neglect
is perhaps a more appropriate term to
use considering the millions of dollars of research and development extended each year to develop these RF
components.
Should we challenge our industry and RF component suppliers
to move quickly to these more dynamic metrics? It seems the prudent
choice. The first step is a straightforward one—change the datasheets
and show the performance under
these new metrics. Only then will
original equipment manufacturers
(OEM) see where improvements can
be made, which will enhance the operational quality of the handsets and
mobile devices they create. We invite
our fellow suppliers to join us as we
send the message—Welcome to the
next decade of RF. 
Ben Thomas is the director of marketing
for 3G/4G Cellular Front Ends at RFMD.
Jackie Johnson is the manager of
Cellular Front End Applications
Engineering at RFMD.