Cubuc metric in 3 gpp lte

3GPP TSG RAN WG1 #44 Tdoc# R1-060385
Denver, USA
February 13-17, 2006

Agenda Item
Source: Motorola
Title: Cubic Metric in 3GPP-LTE
Document for: Discussion

Introduction
The conclusion of the original cubic metric (CM) discussion [1] states the peak-to-average power
ratio (PAPR) of a signal does not predict PA power de-rating as accurately as CM. This paper will
expand on the original conclusions by analyzing certain OFDM-type signals that have been considered to
meet the long-term evolution (LTE) goals for 3GPP networks.

Review of the Cubic Metric
A cubic metric has been adopted by the 3GPP members as a method to determine PA power de-
rating because of its accuracy over a wide range of devices and signals. This method has proven to be
superior, for W-CDMA signals, compared to methods that use the statistical PAPR to predict de-rating.
With the introduction of various OFDM-type modulation formats it is crucial that the cubic metric is
verified as a valid predictor of power de-rating.
Data has been collected on several devices for a variety of signals that will show how well the de-
rating of new LTE signals is predicted by CM. Recall the equation to compute the power de-rating via
the cubic metric method:

CM =
[ ]
20 log10 {rms v norm ( t ) } − 20 log10 {rms vref norm ( t ) }
3 3
[ ] dB
K

Where
[ ]
20 log10 {rms vnorm ( t ) } is the called raw cubic metric (in dB) of a signal
3

[ ]
20 log10 {rms vref norm ( t ) } = 1.52 dB is the raw cubic metric of the W-CDMA voice
3

reference signal

rms ( x ) =
( x′x ) v norm ( t ) =
v(t )
to clarify: N , and rms[ v ( t ) ]

In previous work K was empirically determined to be 1.85 for a set of W-CDMA signals.
A new value for K will be determined for multi-carrier signals in this report.

In words, this equation computes the cubic power of a signal v(t), compares it to a reference
signal vref(t) and uses the empirical slope factor K to complete the estimate. Although the signal vectors
are shown as a function of time, the time scale is effectively removed in the rms operation and is not a
part of the computation. Thus the raw CM result is not a function of symbol/chip rate (or alternatively
not a function of signal bandwidth). This can be easily verified by computing the raw cubic metric for a
conventional W-CDMA signal at a 3.84 MHz chip rate, and for the same signal generated at a 7.68 MHz
chip rate. As long as the SRRC filter bandwidth is scaled with the chip rate, an essentially identical
signal is generated at twice the bandwidth, but having the same raw cubic metric. Similarly, a change in
chip rate has no affect on PAPR. Bandwidth effects will be significant to this study since it includes
signals with 3 dB bandwidths larger than the 3.84 MHz used for W-CDMA.

1

Bandwidth Considerations
Even though raw cubic metric is not affected by signal bandwidth, the adjacent channel leakage
ratio (ACLR) can be, depending on the definition of the ACLR measurement. This is illustrated in Figure
1 below. The solid shaded boxes represent example spectra of a non-distorted signal and its third order
(cubic) distortion products. The dashed-line boxes represent channel and filter bandwidths and offsets.
The channels and filters in a.) are defined by the 3GPP and are used for all W-CDMA standard
measurements. b.) shows that the frequency overlap between the 3GPP defined measurement filters and
the spectrum of the distortion products increases as the signal bandwidth increases. Comparing a.) and b.)
one can see that ACLR, which is integrated across the measurement filter bandwidth, will be higher for
b.). This will be true even if the relative spectral densities between on channel and adjacent channel
spectra are equal for a.) and b.) simply because the measurement filter is collecting adjacent channel
power over a larger fraction of its bandwidth.

Figure 1 – Considerations for occupied bandwidth in adjacent channel power measurements
|S(f)|

• 3.84 MHz transmitter filter bandwidth
a.) • 5 MHz channel separation
• 3.84 MHz channel power integration bandwidth

f

• 4.51 MHz transmit filter bandwidth
b.) • 5 MHz channel separation

f

• 4.51 MHz transmit filter bandwidth
c.) • 5 x 4.51/3.84 MHz channel separation

f

Approximate Signal Spectrum 5 MHz channel 3.84 MHz integrated channel power

Approx . Cubic Spectrum 5 x 4.51/3.84 MHZ channel 4.51 MHz integrated channel power

Many of the proposed LTE signals have an increased occupied bandwidth with the goal of
improving spectral efficiency. Since these bandwidth effects are significant, and it has already been
determined that raw cubic metric cannot capture them, a modified technique must be used to study a
signal space that includes signals of different bandwidths. In this new technique all signal measurements
are first rescaled to an equivalent bandwidth. Next the relationship between raw cubic metric and PA
power capability is determined, just as it was in the original W-CDMA cubic metric work. Finally, an
empirical correction is determined that relates the rescaled equivalent bandwidth measurements back to
measurements made using the actual measurement bandwidths and channel spacings deployed.
The diagrams in Figure 1 suggest a simple technique to rescale all the measurements. In a.) one
can observe that only a certain fraction of the third order distortion spectrum falls within the measurement
filter bandwidth, and also that only a certain fraction of the measurement filter bandwidth intercepts any
of the third order distortion power. In b.) both of those fractions have been altered by the increased signal
bandwidth. What is needed is an adjustment that restores both of those fractions to their original values
for the wider bandwidth signal. The adjustment shown in c.) achieves this goal by increasing both the
measurement filter bandwidth and channel separation by the same multiplier. That multiplier is the ratio
of the 3 dB bandwidths being considered. In c.) the study of a 4.51 MHz bandwidth signal is carried out
using a measurement bandwidth of (4.51/3.84) x 3.84 MHz, and a channel separation of

2

(4.51/3.84) x 5 MHz. Note that this is for study purposes only. Later in the development results will be
provided for a conventional configuration with 5 MHz channels and 3.84 MHz measurement bandwidths.

Methodology
Several proposed LTE uplink signals with varying levels of PAPR and CM are generated. These
are described in Table 1 below.

Table 1 – Descriptions of LTE Signals Tested
CP Raw CM
Signal Sys Map NFFT Fract NActive Modn BW (MHz) (dB)
A OFDM* PUSC-UL 512 0.25 408 16-QAM 4.51 7.75
B DFTS-OFDM UL 512 0.0625 300 QPSK 4.51 3.44
C DFTS-OFDM UL 512 0.0625 300 16-QAM 4.51 4.85
D DFTS-OFDM UL 512 0.0625 300 64-QAM 4.51 5.18
E IFDMA Full 512 0.25 512 QPSK 3.84 2.40
F IFDMA Full 512 0.25 512 16-QAM 3.84 4.36
G IFDMA Full 512 0.25 512 64-QAM 3.84 4.64
*Results from QPSK and 64-QAM modulations for the OFDM signal are identical to results from 16-QAM modulation; the three
cases can be adequately represented by signal ‘A’ alone

Note that signal ‘A’ is the only modulation of the OFDM type (QPSK and 64QAM have been
omitted). Simulations show that there is no change in PAPR or CM between different modulation
formats of the OFDM system; measurements confirm that there is no power de-rating between the
formats. Only the one modulation case is analyzed here in an effort to avoid artificially weighting data at
any particular CM or PAPR value.
These signals are applied to three different power amplifiers designed for UE applications.
Different PA technologies are represented by two GaAs HBT devices and one GaAs EpHEMT device.
All are 50 ohm matched power amplifier modules representative of current technology.

Table 2 – Description of Power Amplifiers Tested
PA # Description
1 GaAs HBT
2 GaAs HBT
3 GaAs EpHEMT

Each of the 3 power amplifiers are tested with each of the 7 LTE signals and the W-CDMA voice
signal for reference. The input power is swept and the output power level at which the adjacent channel
(± 5 MHz) ACLR reached -33 dB (the linear power capability or LPC) is recorded for each PA and signal
combination1. A rescaled bandwidth-equivalent adjacent channel offset (4.51/3.84 x ± 5MHz = ±
5.87MHz) and measurement bandwidth (4.51/3.84 x 3.84MHz = 4.51MHz) is used for measuring signals
‘A’ through ‘D’ to remove effects of the higher signal bandwidth. Standard measurements using ± 5
MHz channel offsets and 3.84 MHz measurement filter bandwidths are also performed for this set of
signals to determine the bandwidth de-rating component once the raw CM component is known. The
power capability is compared with the W-CDMA signal to determine the relative LPC, or power de-
rating, necessary for each signal and PA combination. In this way, the absolute power capabilities of the
three PA's tested is immaterial; only the signal characteristics affect the results.

1 Note that -33 dB ACLR has been chosen which is consistent with W-CDMA and serves as useful reference value for LTE analysis until
WG4 decides whether ACLR requirements should be changed.

3

Results and Analysis
Table 3 shows the output power de-rating for each signal and each device.

Tables 3 & 4 – Measured Relative Power Capability by Signal and PA
Bandwidth Normalized Results Standard Results
Power De-rating Power De-rating
Signal Raw CM PA #1 PA #2 PA #3 Signal Raw CM PA #1 PA #2 PA #3
Ref. 1.52 - - - Ref. 1.52 - - -
A 7.75 3.51 4.08 3.99 A 7.75 4.46 4.93 4.87
B 3.44 1.34 1.63 1.73 B 3.44 2.14 2.29 1.85
C 4.85 1.92 2.26 2.28 C 4.85 2.75 2.89 2.43
D 5.18 2.12 2.44 2.59 D 5.18 3.06 3.07 2.71
E 2.40 - - - E 2.40 0.48 0.45 0.63
F 4.36 - - - F 4.36 1.50 1.74 1.84
G 4.64 - - - G 4.64 1.79 2.03 1.98
(3) (4)

The relationship between the amount of power de-rating and the raw cubic metric is shown below
in Figure 2; the relationship determines the final cubic metric equation (without bandwidth effects). The
points for this plot are taken from Table 3 for signals ‘A’ through ‘D’ and Table 4 for signals E through
G. This corresponds to selecting bandwidth normalized results for the 4.51 MHz bandwidth signals and
standard results for the standard 3.84 MHz bandwidth signals. A linear-best-fit regression line is
calculated and plotted among the data points. The slope of this line quantifies K from the cubic metric
equation for this set of signals. K for this set of signals is different from previous experiments using only
W-CDMA signals (K = 1.56 for LTE vs. K = 1.85 for W-CDMA). Standard error, σy, is 0.23 dB. This
term is the standard deviation of the error between measured data points and the regression line.
To show that CM is superior to PAPR in de-rating prediction, power de-rating vs. the increase in
PAPR over the voice case is plotted in Figure 3. A line with a slope of 1 and passing through the
reference W-CDMA data point, corresponding to predicting the de-rating directly from the increase in
PAPR above the reference signal, is also plotted. Standard error of PAPR prediction is 1.56 dB, on
average more than 1.33 dB poorer than CM prediction.
As described earlier, an empirical correction must be determined for signals ‘A’ through ‘D’ that
relates the rescaled equivalent bandwidth measurements back to measurements made using the standard
method (where actual measurement bandwidths and channel spacings are deployed). The average
difference between the bandwidth normalized measurements and standard measurements is shown in
Figure 4. The average difference, 0.77 dB, is applied as an offset to the regression line from Figure 2.
This offset will be used to modify the CM equation to account for signals of different occupied
bandwidths. One implication of this result is that there is a penalty for increasing spectral efficiency
(increasing signal bandwidth without modifying channel spacing). Methods to mitigate the bandwidth
offset will be discussed in a related report [2].
While this empirical offset only applies to the set of signal and filter bandwidths and offsets
studied in this paper, it is worth noting that the ratio of bandwidths 4.51/3.84 is equal to 0.70 dB. The
proximity of this theoretical value to the empirical 0.77 dB result is probably not a coincidence suggests
an area for further study.

4

Figure 2
Power De-rating Prediction Using Cubic Metric

4.50

4.00
σy = 0.227 dB
3.50

3.00 PA #1
Power De-rating (dB)

PA #2
2.50

PA #3
2.00
CM Regression Line
(slope = 1/1.56)
1.50

1.00

0.50

0.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Raw Cubic Metric - 1.52 (dB)

Figure 3
Power De-rating Prediction Using PAPR

4.00

3.50

σy = 1.56 dB
3.00
PA #1

2.50

PA #2

2.00
PA #3

1.50 PAPR Voice Ref. Line
(slope = 1)

1.00

0.50

0.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00
PAPR - 2.92 (dB)

5

Figure 4
ACLR Filter Offset Adjustment
For Bandwidth Normalized Signals

5.00

4.50

4.00

5.0 MHz Filter Offset
3.50

3.00 5.87 MHz Filter Offset

2.50
CM Regression Line
(slope = 1/1.56)
2.00

Offset CM Regression
1.50 Line (offset = 0.77 dB)

1.00

0.50

0.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
Raw Cubic Metric - 1.52 (dB)

Conclusion
• Results show that the prediction accuracy of CM is superior to that of PAPR.
• Measuring the higher bandwidth signals (‘A’ through ‘D’) with scaled adjacent channel
definitions does not give the actual de-rating in real system performance (measurement with
system defined adjacent channel offset). A de-rating increase of 0.77 dB from the CM regression
line can be used to estimate the actual de-rating of higher bandwidth signals. (Note this empirical
offset only applies to the set of signal and filter bandwidths and offsets studied in this paper).
• The amount by which the power capability of a UE power amplifier must be de-rated for LTE
signals with 4.51MHz nominal bandwidth can be summarized by equation (1):

CM =
3
[
20 log10 {rms vnorm ( t ) } − 1.52 ]
+ 0.77 dB (1)
1.56
where 0.77 dB is the increase in CM due to the change in signal bandwidth relative to 3.84 MHz,
if there is no change in relative bandwidth then equation 2 below applies

• The amount by which the power capability of a UE power amplifier must be de-rated for LTE
signals with 3.84MHz nominal bandwidth can be summarized by equation (2).

CM =
3
[
20 log10 {rms vnorm ( t ) } − 1.52]dB (2)
1.56

• The final CM for each signal is shown in Table 5 below and is plotted against measured power
de-rating (using standard measurement method results from Table 4) in Figure 5. The standard
error of the final estimate is 0.187 dB.

6

Table 5 – Final CM Summary
Signal Bandwidth
Type Modn Raw CM (dB) K Offset (dB) CM (dB)
WCDMA voice 1.52 - 0.00 0
OFDM 16-QAM 7.75 1.56 0.77 4.76
DFTS-OFDM QPSK 3.44 1.56 0.77 2.00
DFTS-OFDM 16-QAM 4.85 1.56 0.77 2.90
DFTS-OFDM 64-QAM 5.18 1.56 0.77 3.11
IFDMA QPSK 2.40 1.56 0.00 0.56
IFDMA 16-QAM 4.36 1.56 0.00 1.82
IFDMA 64-QAM 4.64 1.56 0.00 2.00

Figure 5
Final CM For LTE Signals

5

4.5

σy = 0.187 dB
4

3.5

3

2.5 Measured Data
(standard method)

2 CM Power De-
rating Prediction

1.5

1

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
CM (dB)

References
[1] R1-040642, “Comparison of PAR and Cubic Metric for Power De-rating”, 3GPP RAN WG1 #37, Montreal 2004.
[2] R1-060144, “UE Power Management for E-UTRA”, 3GPP RAN WG1 LTE Adhoc, Helsinki 2006.

7

Cubuc metric in 3 gpp lte

More Related Content

What's hot (20)

Viewers also liked (15)

Similar to Cubuc metric in 3 gpp lte (20)

Recently uploaded (20)

Cubuc metric in 3 gpp lte