nonlinear equalization processor ic for wideband receivers
TRANSCRIPT
Nonlinear Equalization Processor qIC for Wideband Receivers and
SensorsSensorsWilliam S. Song, Joshua I. Kramer, James R. Mann,
Karen M. Gettings, Gil M. Raz, Joel I. Goodman, a e Gett gs, G a , Joe Good a ,Benjamin A. Miller, Matthew Herman, Thomas B.
Emberley, Larry L. Retherford, Albert H. Horst
HPEC 2009HPEC 2009
23 September 2009RESEARCH & TECHNOLOGY, INC.
This work was sponsored by DARPA under Air Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions
HPEC 2009-1WSS 9/23/2009
MIT Lincoln Laboratory
This work was sponsored by DARPA under Air Force Contract FA8721 05 C 0002. Opinions, interpretations, conclusions and recommendations are those of the authors and are not necessarily endorsed by the United States Government.
Outline
• Introduction– Nonlinear Equalization (NLEQ) applications– Nonlinear Equalization (NLEQ) applications– NLEQ program objective
• NLEQ processor architecture• VLSI NLEQ processors• VLSI NLEQ processors• Performance demonstration results• Summary
MIT Lincoln LaboratoryHPEC 2009-2
WSS 9/23/2009
High Dynamic Range Requirements for Military and Commercial Sensor Systems
Radar Receiver System Example Radar Receiver System
JammingRadar
Radar Receiver System Example Radar Receiver SystemInterference + Noise
Environment
B)
Clutter
Jammer
Dynamic Range
Target
Pow
er (d
B
Noise
g80->100 dB
Range
Small targets at or below noise level
• Radar, ELINT, SIGINT, Comm receiver systems must support high dynamic range operation
– To detect small targets/signals in interference/clutter environment
MIT Lincoln LaboratoryHPEC 2009-3
WSS 9/23/2009
To detect small targets/signals in interference/clutter environment– High signal-to-noise ratio and linearity required
Linearity Concerns in Highly Digitized Arrays and Frequency Channelized Systems
• Digital beamforming and frequency channelization increase in-band signal-to-noise ratio (SNR)Receiver ADCReceiver ADCReceiver ADC
Digital Sensor Array
• Non-linearity generated spurs and intermods can interfere with small signal detection
ece e CReceiver ADCReceiver ADCReceiver ADCReceiver ADCReceiver ADCReceiver ADC
Noise
Two-toneinput
Inter-
Pow
er Two-toneinput
Inter-mods
Pow
er Two-toneinput
Inter-mods
Pow
er
mods
Frequency
Noiseods
Frequency
Signal
Noiseods
Frequency
Receiver ADC FrequencyChannelizer Po
wer
Out-of-Band Noise
In-Band Noise
MIT Lincoln LaboratoryHPEC 2009-4
WSS 9/23/2009
Frequency FNyquist
NoiseFrequency Channelized Receiver
Nonlinear Equalization (NLEQ)
Rec A/DFrom Antennas
1
N
Digital ReceiverDigital Receiver
DigitalSignal
LNAAntennas
Sources of Nonlinearities
gProcessor
MIT Lincoln LaboratoryHPEC 2009-5
WSS 9/23/2009
Nonlinear Equalization (NLEQ)
Rec A/D
1
N
Digital ReceiverDigital Receiver
DigitalSignal
LNA
Sources of Nonlinearities
gProcessor
In-Band Intermodulation Distortions
Before Equalization
r (dB
)Po
wer
Frequency (bin)
MIT Lincoln LaboratoryHPEC 2009-6
WSS 9/23/2009
Nonlinear Equalization (NLEQ)
Rec A/DFrom Antennas
1
N
Digital ReceiverDigital Receiver
DigitalSignal
Processor
NonlinearEqualizer Digital
SignalLNA
Antennas
Sources of Nonlinearities
Processor
Inverts Nonlinearities
SignalProcessor
In-Band Intermodulation Distortions
Before Equalization
r (dB
)
After Equalization
(dB
)
Pow
er
Pow
er (
Dynamic range improvement
Frequency (bin) Frequency (bin)
MIT Lincoln LaboratoryHPEC 2009-7
WSS 9/23/2009
Nonlinear Equalization (NLEQ)
Rec A/DFrom Antennas
1
N
Digital ReceiverDigital Receiver
DigitalSignal
DigitalSignal
Processor
NonlinearEqualizer Digital
Signal
+20~25 dB SFDR/IFDR
LNAAntennas
Sources of Nonlinearities
gProcessor Processor
Inverts Nonlinearities
SignalProcessor @500MHz BW
>1 Teraops< 2 Watts
In-Band Intermodulation Distortions
Before Equalization
r (dB
)
After Equalization
(dB
)
Pow
er
Pow
er (
Dynamic range improvement
Frequency (bin) Frequency (bin)
• Nonlinear equalizer processor can reduce nonlinear distortion levels in analog and mixed signal circuitry
MIT Lincoln LaboratoryHPEC 2009-8
WSS 9/23/2009
g g y• Equivalent to having devices 10-20 years ahead of their time
Outline
• Introduction• NLEQ processor architecture• NLEQ processor architecture
– Processor architecture– Architecture optimization
• VLSI NLEQ processors• VLSI NLEQ processors• Performance demonstration results• Summary
MIT Lincoln LaboratoryHPEC 2009-9
WSS 9/23/2009
NLEQ Polynomial Nonlinear Filter Architecture
Z-α Pass-through Σ
Pass-through
Processing Elements
Z β XFIR
• Parameters to optimize:– Number of processing
elements
Z-β XFIRZ-β XFIR
Z-β
Z-γ1
XFIR ΣZ-β
Z-γ1
XFIR elements– Polynomial orders of
processing elements– Number of taps
Z-γ1Z-γ1
Z
Z γ1
Z-γΝZ-γΝ
– Delay values
Z-γΝZ-γΝ
MIT Lincoln LaboratoryHPEC 2009-10WSS 9/23/2009
Wideband NLEQ Processor Parameter Optimization
Z-α Pass-through
Z-β XFIR
Z-γΝ
Z-γ1
Z-γΝ
• Parameter selection
MIT Lincoln LaboratoryHPEC 2009-11WSS 9/23/2009
– 10 filters, 10 taps, 0-3 tap delay, 0-3 group delay, 3 pass through delay
Parallel NLEQ Processor Architecture Optimization
P P P P e
Distributed Polyphase Block-Floating-Point Residue Arithmetic Architecture
Low VT Dynamic Logic IC Layout
P P P P
P P P P
P P P P
Dis
trib
ute
Rea
ssem
ble
Design Feedback(Performance Trades)
Design Feedback(Performance Trades)
Sub-processor Task Distribution & Sizing
FIR1φ1
IC Architecture and Floor Plan
FIR1φN NLP1φ1 NLP1φN
P P P P P P P P P P P P
P P P P P P P P P P P PFIR1φ1 FIR1φ2 FIR1φ3 I/OClocksDesign
FIR1φ2 FIR2φN NLP2φ1 NLP2φN
FIRmφ1 FIRmφN NLPmφ1 NLPmφN
P P P P P P P P P P P P
P P P P P P P P P P P P
P P P P P P P P P P P P
P P P P P P P P P P P P
NLP1φ1 NLP1φ2 NLP1φ3
FIR2φ1 FIR2φ2 FIR2φ3Feedback
(Routing Efficiency)
Buses
MIT Lincoln LaboratoryHPEC 2009-12WSS 9/23/2009
Memory ControlEfficiency)
Measured Results: PRN Signal with Amplifier & ADC Distortions
TypicalUnequalized Equalized
In-band i d
Typical Response
intermods Harmonics
wer
(dB
FS)
wer
(dB
FS)
Pow
Pow
• Non-tonal signal experiment parameters– Amplifier in saturation region
M 104 i l d d i t tb d t– Max 104 included in testbed system– Pseudo-Random Noise (band-limited) input waveform
• Greater than 25 dB dynamic range improvement– In-band signal component is not distorted by the equalizer
MIT Lincoln LaboratoryHPEC 2009-13WSS 9/23/2009
In band signal component is not distorted by the equalizer
MIT Lincoln Laboratory
NLEQ Adaptive Equalization Performance
Adaptive Equalization PerformanceFully Adaptive
Partially Adaptive
Non-Adaptive-14
-12dB
Adaptive Equalization Performance
-18
-16
uctio
n in
em
ent (
dB)
-22
-20
ur R
edud
uFD
R Im
prov
e
Minimal performancelosses uses computationallyffi i t d ti
0 10 20 30 40 50-26
-24SpuIF efficient adaptive
algorithmNominal training temperature
Ambient Temperature in Degrees Celcius• Characterized Max 108 device sensitivity to changes in temperature• Applied adaptive equalizations
Fully temperature adaptive approach achieve good performance
Ambient Temperature (oC)
MIT Lincoln LaboratoryHPEC 2009-14WSS 9/23/2009
– Fully temperature adaptive approach achieve good performance– Partially adaptive approach can achieve similar performance with much
higher computational efficiency
Outline
• Introduction• NLEQ processor architecture• NLEQ processor architecture• VLSI NLEQ processors
– Enabling technologiesNLEQ4000 ultra wideband processor IC– NLEQ4000 ultra-wideband processor IC
– NLEQ500 wideband processor IC• Performance demonstration results• Summary• Summary
MIT Lincoln LaboratoryHPEC 2009-15WSS 9/23/2009
Nonlinear Equalizer Processor IC Development Process
Non-Linear EqualizerAlgorithm/Architecture
EQ FIRPartition FIR #1
×
H1
H2 +Partition FIR #2Distributed Polyphase IC ArchitecturePartition FIR #N
HJ
X(n) Y(n)
P P P P
P P P P
P P P P
P P P PDis
trib
utio
n
Rea
ssem
ble
ypResidue Arithmetic
FIR1φ1 FIR1φNNLP1φ1NLP1φNFIR1φ2 FIR2φNNLP2φ1NLP2φN
FIRmφ1 FIRmφNNLPmφ1NLPmφN
I/OClocks
Buses
NLEQ Signal Processor ICP P P PD R Memory Control
● 1.5 Tearops● 1.2 Watts
8
161
2 2
4
816
Optimal RegionDegree of Polyphasing
Throughput DensityPower Efficiency
Architecture Optimization
500 MH BW MAX108
Low Vt Dynamic Logic
124
8
4816
1
2
Level of Pipeline
EfficiencyThroughput Density Times Power Efficiency
500 MHz BW MAX108
Pow
er (d
B) Before NLEQ
After NLEQ
MIT Lincoln LaboratoryHPEC 2009-16WSS 9/23/2009
Frequency Bin
Architectural Comparison of Pentium 4 and MIT LL Processor Dies
Pentium 4 Die MIT LL Processor Die
ProcessorArray
• Single processor• Multiple caches and memory bus
• 1000’s of parallel processors• Local memory only• Multiple caches and memory bus
• Only core runs at high speed• General purpose processing• Large design team (>100)
• Local memory only• Entire die runs at high speed• Signal processing functions• Small design team (<10)
MIT Lincoln LaboratoryHPEC 2009-17WSS 9/23/2009
g g ( ) g ( )
Custom CMOS Circuit Design
Standard Cell Full Adder Full Custom Full Adder
• Custom logic and devices sizing• Manual device placement and
interconnect• Small area and power
consumption
• Pre-designed logic gates only• Automatic device placement and
routing• Large area and power
consumption
MIT Lincoln LaboratoryHPEC 2009-18WSS 9/23/2009
consumptionconsumption
Full Custom Low Threshold Voltage (VT) Dynamic Logic Circuits
Standard Cell Static Register Full CustomDynamic Register
Low VT Devices
Charge
Low VT Devices
P = ½CV2f
Leakage
● Frequent refresh● Bypass capacitors
Design Techniques
● Bypass capacitors● Signal isolation● Guard rings● Robust circuits
• Logic based computation and storage
• Many devices
• Charge based computation and storage
• Fewer devices
MIT Lincoln LaboratoryHPEC 2009-19WSS 9/23/2009
• Large area and power consumption • Small area and power consumption
NLEQ4000 Processor IC
• Up to 4,000MSPS• Selectable bit widths
– Up to 12 bits input
Wideband NLEQ Processor IC LayoutAnd BGA Package Up to 12 bits input
– Up to 16 bits output• LVDS and CMOS I/O• Programmable coefficients
And BGA Package
g• Block floating point residue
arithmetic
Parameter(IBM 0.13um)
Core Chip
Die Size 2.6 mm x 3.3 mm
6mm x 6mm
Power @1500MSPS
266mW 453mW
Power @4000MSPS
706mW 1219mW
MIT Lincoln LaboratoryHPEC 2009-20WSS 9/23/2009
NLEQ500 Processor
• Up to 500MSPS• Selectable bit widths
– Up to 18 bits input
IBM 0.13µm Die
– Up to 18 bits input– Up to 22 bits output
• Low voltage CMOS I/O• Programmable configuration and coefficientsg g• Block floating point residue arithmetic• Yield 14/15 for LowVt and 14/15 for RegVt
Parameter Core I/O ChipParameter(IBM CMOS 0.13um)
Core I/O Chip
Size 0.65mm x 1.4mm
N/A 2.2mm x 2.2mm
P @100MSPS 6 W 19 W* 25 W*
BGA
Power @100MSPS(Vdd=0.6V)
6mW 19mW* 25mW*
Power @500MSPS(Vdd=1.2V)
122mW 121mW* 243mW*
MIT Lincoln LaboratoryHPEC 2009-21WSS 9/23/2009
*With 50 Ohm Load Termination
Outline
• Introduction• NLEQ processor architecture• NLEQ processor architecture• VLSI NLEQ processors• Performance demonstration results
S• Summary
MIT Lincoln LaboratoryHPEC 2009-22WSS 9/23/2009
MAX108 Demonstration ResultsWithout NLEQ4000Without NLEQ4000
MAX108 ADC with NLEQ4000
NLEQ4000
With NLEQ4000
NLEQ4000MAX108
• ~21 dB linearity
Q
improvement demonstrated with NLEQ4000 IC at 1.5GSPS
MIT Lincoln LaboratoryHPEC 2009-23WSS 9/23/2009
Measured NLEQ Performance Improvement for Other ADCs
Interleaving Architecture~1 GHzImproved ~17 dB
Folding Architecture1.5 GHzImproved ~13 dB
National 81000 Atmel AT84AS008 Flash ArchitectureFlash Architecture~1.3 GHzImproved ~12 dB Entire Receiver Chain
30 MHz BWImproved ~12 dB
MIT Lincoln LaboratoryHPEC 2009-24WSS 9/23/2009
Maxim 109 LTC 2209
X-Band Receiver-on-Chip (RoC) Development Based on NLEQ DSP
BPF BPFLO
ADC NLEQ
Top View
MAX108
ADC NLEQRoC
Bottom View
LNA IF Amp
ADC
Bottom View
Mixer Integrated RoC Die360mW
• Linear dynamic range limited by the final IF amplifierNLEQ
360mW
Linear dynamic range limited by the final IF amplifier– NLEQ DSP to linearize the amplifier and ADC
• High performance and low power achieved with new analog/digital co-design paradigm
MIT Lincoln LaboratoryHPEC 2009-25WSS 9/23/2009
• Single die receiver implementation being explored
Summary
• Linearity enhancement required by DoD/commercial sensor/receiver applications
– Phased array sensors/receivers– Frequency channelized sensors/receivers
• MIT LL has developed high-throughput low-power nonlinear equalization signal processor ICs
– Massively parallel systolic architecturey p y– Polyphase distributed arithmetic processing– Block floating point residue number arithmetic– Full custom low-threshold-voltage dynamic logic
• Successful demonstration results• Successful demonstration results– >20 dB linearity improvement– NLEQ4000
Up to 4GSPS, <1.25WUp to 12 bit ADCsUp to 12 bit ADCs
– NLEQ500Up to 500MSPS, <0.25WUp to 18bit ADCs
MIT Lincoln LaboratoryHPEC 2009-26WSS 9/23/2009