Title: Texas Instruments Linearization Fundamentals Driving Digital Pre-Distortion and the GC5322!
1Texas InstrumentsLinearization Fundamentals
Driving Digital Pre-Distortionand the GC5322!
April 2006
2Agenda
- Introduction and Impact
- Origin and History of the Problem
- Linearization Fundamentals
- Polynomial Power Amplifier Modeling
- Crest Factor Reduction
- Digital Pre-Distortion
- System Implementation
- Crest Factor Reduction and Digital Pre-Distortion
- Adaptive Memory Pre-distortion of Power
Amplifiers - Conclusions
3Introduction and Impact
- The demands for spectrally efficient modulation
schemes have increased however these schemes are
subject to severe intermodulation distortion
(IMD) when the power amplifiers (PA) are operated
near saturation - Unfortunately, PAs are most efficient when
operated near saturation
Super 3G 4G
20MHz BW_at_ 2.1GHz
WiMAX .16a/d/e
10-40MHz BW _at_ 2.5, 3.5 5GHz
WCDMA
5MHz BW_at_ 2.1GHz
Cellular Channel BW _at_ Band
Increased signal bandwidth and complexity
CDMA 2000
1.25MHz BW _at_ 1.9GHz
A big challenge forMCPA designers!
EDGE/ CDMA
200kHz BW _at_ 800MHz
TDMA/ GSM
lt200kHz BW _at_ 8-900MHz
AMPS/ D-AMPS
30kHz BW _at_ 800MHz
4Introduction and Impact
- High Power RF PAs (gt10W) use multiple driver
stages to amplify an input signal. - Different PA architectures (Class A, AB, C, etc
) offer various degrees of linearity, cost and
efficiency. - RF PAs are notoriously inefficient Air is a
convenient but poor transmission medium. - RF PAs are designed (tuned) for specific
frequency range and bandwidth - MCPA wideband RF PA, does not have to process
multiple carriers - PA Gain is usually fixed so pre-amps may be
required to drive the PA input.
TX Board
Antenna
RFout 50dBm(100W)
RFin 20dBm_at_ gt800MHz
FromBaseband
PA
DAC
IF-gtRF
DUC
A
50 OhmTypicalInput
3 to 4 gain stages typical
If Gain 30dB,
Pre-Amp
5Introduction and Impact
- Linearization techniques allow a PA to be
operated at higher power with minimal IMD
increases, thus greater efficiency - Recent technological advances have made digital
pre-distortion the focus of research efforts - Crest factor reduction (CFR) further increases
the efficiency of the PA by reducing the
peak-to-average ratio (PAR) of the transmitted
signal
Theoretical Performance of Class AB PA
Pre-Distortion No No Yes Yes
CFR No Moderate Moderate Yes
Tx Power 10W 10W 10W 10W
PAR 12dB 9dB 9dB 6dB
Backoff 15dB 12dB 9dB 6dB
PA Power Rating 320W 160W 80W 40W
Efficency 5 9 18 30
Power Dissipation 120W 101W 45W 7W
6Origin and History of the Problem
1. Linearization Fundamentals
- The trade-off between efficiency and linearity is
the primary concern for PA designers - A PA operating at a high percentage of its power
rating requires external linearization to
maintain linearity - The linearization of the PA reduces back-off,
thus increasing efficiency
7Origin and History of the Problem
2. Polynomial Power Amplifier Modeling
- Accurate representation of the nonlinear effects
in PAs is achieved using a polynomial expression,
as follows - The coefficients represent the linear gain, and
the gain constants for the quadratic and cubic
nonlinearities - A system with memory (phase) versus memory
effects (non-linearities) - Envelope and frequency memory effects
8Origin and History of the Problem
2. Power Amplifier Characterization
- Two tone test is useful for measuring spectral
regrowth in a nonlinear and memoryless system
9Origin and History of the Problem
2. Power Amplifier Characterization
- Theoretically, only odd-degree nonlinearities
generate in-band distortion products - The simplified polynomial PA model is expressed
as follows
10Origin and History of the Problem
2. Power Amplifier Characterization
- A PA is often characterized by its
amplitude-amplitude and amplitude-phase transfer
characteristics - The simple polynomial is unable to model AM-PM
effects - Both AM-AM and AM-PM effects are represented by
the complex baseband model
where
11Origin and History of the Problem
2. Power Amplifier Characterization
- A simple case considering only 3rd degree
nonlinearities in the AM-AM and AM-PM transfer
characteristics is represented by the following - In the linear range, the PA can be characterized
by the following
and
12Origin and History of the Problem
2. Power Amplifier Characterization
AM-AM Characteristic
AM-PM Characteristic
13Origin and History of the Problem
3. Crest-Factor Reduction
- The DPD optimal performance depends greatly on
signal characteristics - Multi-carrier signals can have a PAR as high as
13dB increasing the level of back-off to maintain
acceptable IMD levels - The application of CFR allows the PA to operate
at higher input/output power levels while
maintaining linearity at the output of the PA - Achieved through pulse generation and digital
clipping
14Origin and History of the Problem
3. Crest-Factor Reduction
- Preferred PA bias point for a typical modulated
signal
15Origin and History of the Problem
3. Crest-Factor Reduction
- Preferred PA bias point for a CFR signal
16Origin and History of the Problem
4. Digital Pre-Distortion
- Pre-distortion effectively performs a
mathematical inversion of the Volterra PA model - The output of the pre-distortion processor is
described by the following - The PA is linearized when
17Origin and History of the Problem
4. Digital Pre-Distortion
- Digital pre-distortion (DPD) has become an
effective linearization technique due to the
renewed possibilities offered by DSP - Adaptive PD designs use feedback to compensate
for PA variations - Look-up tables are updated to achieve optimal
pre-distortion by comparing PD input to PA output - The PD function is expressed as a complex
polynomial
where
18Origin and History of the Problem
4. Digital Pre-Distortion
- Digital pre-distortion (DPD) requires feedback
for sample-by-sample adaptation 5 times that of
the signal bandwidth - Multi-carrier systems use signal bandwidths of up
to 20MHz today, thus the feedback bandwidth must
be 100MHz to compensate 3rd and 5th order IMD - Least-mean-square (LMS) is a popular gradient
based optimization algorithm that requires
wideband feedback
19System Implementation
1. Crest-Factor Reduction and Digital
Pre-Distortion
- The combination of CFR and digital pre-distortion
were investigated - In this case, linearization was achieved with a
traditional wideband feedback LMS algorithm - The CFR technique used was proposed by Texas
Instruments using the GC1115 signal pre-processor - Four stages ensure that the output PAR is reduced
to values from 5 to 8dB, as specified by the user - Performance results were compared using a Cree
Microdevices 30W PA operating at 1.96GHz and a
signal bandwidth of 1.25MHz - The PAR of the IS-95 signal was reduced from
9.6dB to 5dB
20System Implementation
1. Crest-Factor Reduction and Digital
Pre-Distortion
Complex Canceling Pulse
21System Implementation
1. Crest-Factor Reduction and Digital
Pre-Distortion
Corrected and uncorrected signal with canceling
peaks and detection threshold
22System Implementation
1. Crest-Factor Reduction and Digital
Pre-Distortion
Typical Peak Detection and Cancellation through
Pulse Injection
Cancellation Signal
Input Signal
Output Signal
-
23System Implementation
1. Crest-Factor Reduction and Digital
Pre-Distortion
- Hardware Implementation of Wideband
Pre-Distortion
Waveform Generator
Down-Converter
Attenuator
Analog RF
Agilent 4432B
20dB
DUT
Pre-Distorted Input Signal
LO
Tektronics TDS224 Oscilloscope
Analog IF
24System Implementation
1. Crest-Factor Reduction and Digital
Pre-Distortion
ACPR improvement with respect to output power
25System Implementation
1. Crest-Factor Reduction and Digital
Pre-Distortion
- The ACPR measurements were recorded according to
specifications with a 30kHz marker at and offset
of 885kHz - Results were limited by the performance
limitations of the test bed
Power and efficiency improvement
26System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- The term memory effects refer to the
bandwidth-dependant nonlinear effects often
present in PAs. - These encompass envelope memory effects and
frequency response memory effects. - Envelope memory effects are primarily a result of
thermal hysteresis and electrical properties
inherent to PAs. - Frequency memory effects are due to the
variations in the frequency spacing of the
transmitted signal and are characterized by
shorter time constants.
27System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Memory Polynomial Pre-Distortion Implementation
Where (K7)
And (D2)
28System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Simulated Performance of Wideband Pre-Distortion
- This traditional approach uses and LMS algorithm
to adapt the PD coefficients on a
sample-by-sample basis. - The memory PA model has D1 (delay) and K5
(order).
29System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Simulated Performance of Wideband Pre-Distortion
- The memory PA model is characterized by the
following AM-AM and AM-PM curves
30System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Simulated Performance of Wideband Pre-Distortion
- DPD 0 the LMS algorithm indicates an ACPL
improvement of -3dB and an ACPH improvement of
3dB. - DPD 1 the LMS algorithm indicates an ACPL
improvement of -15dB and an ACPH improvement of
-11dB.
31System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Simulated Performance of Wideband Pre-Distortion
- DPD 2 the LMS algorithm indicates an ACPL
improvement of -24dB and an ACPH improvement of
-23dB. - DPD 3 the LMS algorithm indicates an ACPL
improvement of -24dB and an ACPH improvement of
-20dB.
32System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Hardware Implementation of Wideband
Pre-Distortion - TI offers the complete high-performance signal
chain including DAC5687, CDCM7005, TRF3761,
ADS5444, and TRF3703.
33System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Typical Doherty Amplifier configuration and
Performance Results
34System Implementation
2. Adaptive Memory Pre-distortion of Power
Amplifiers
- Hardware Implementation of Wideband
Pre-Distortion
35Conclusions
- CFR improves DPD performance
- CFR uses EVM and ACLR to tradeoff for added
efficiency - Depending on modulation schemes the relative
percentages may vary - OFDM modulations are sensitive to EVM
- 3GPP modulations are sensitive to ACLR
3GPP Relative Tradeoffs
OFDM Relative Tradeoffs
EVM
ACLR
EVM
ACLR
Efficiency
Efficiency
36Conclusions
- Relative to a PA that operates normally under
backoff, DPD adds additional hardware (cost) and
system complexity to tradeoff for added
efficiency - DPD can effectively remove the negative effects
of CFR enabling even greater levels of efficiency
DPD Relative Tradeoffs
Cost
Complexity
DPD
EVM
ACLR
CFRDPD
CFRDPD
Efficiency
37Questions