Presented by: Shreyas Sen June 11, 2008. Paper 27.3 , DAC 08.

Pro-VIZOR: Process Tunable Virtually Zero Margin Low Power Adaptive RF for Wireless Systems

Pro-VIZOR: Process Tunable Virtually Zero Margin Low Power Adaptive RF for Wireless Systems

Presented by: Presented by: Shreyas Shreyas SenSenJune 11,June 11, 2008. Paper 2008. Paper 27.327.3, DAC , DAC 08.08.

Shreyas Sen, Vishwanath Natarajan, Rajarajan Shreyas Sen, Vishwanath Natarajan, Rajarajan Senguttuvan,Senguttuvan,

Abhijit ChatterjeeAbhijit Chatterjee

Georgia Institute of TechnologyGeorgia Institute of Technology

Goal

Traditional wireless circuits

Designed for worst case environmental variations

and

worst case process corners.

Involves significant built in design margin.

MotivationMotivation

Waste of performance (i.e. power) under favorable condition.

SAVE POWER by adapting to environment and process.

Process Tunable Virtually Zero Margin RFProcess Tunable Virtually Zero Margin RF

BasebandReceiver

Transmitter

BasebandTransmitter

Receiver

Base Station (BS)Base Station (BS) Mobile Station (MS)Mobile Station (MS)

Save Power

Save Power

Worst Case System Design

Worst Case System Design

Adaptive System Design

Adaptive System Design

Varyingchannel

Process adaptability

OutlineOutline

Environment Adaptability in Transceivers

- Adaptive RF receivers. - Adaptive RF transmitters.

Process Adaptability -Process Sensing Using Test.

-Tuning for process variation.

Experimental Results

OutlineOutline

Environment Adaptability in Transceivers

- Adaptive RF receivers. - Adaptive RF transmitters.

Process Adaptability -Process Sensing Using Test.

-Tuning for process variation.

Experimental Results

VIZOR: ReceiverVIZOR: Receiver

Adaptation metric is a measure of the quality of the received signal. Adaptation metric is a measure of the quality of the received signal.

010010

Baseband&RF

Baseband&RF

010010

Transmission ReceptionChannel

BER =tR

Ne

Modulation Demodulation

Ne = Number of errorsR = Data ratet = Test time

Adaptation metric selectionAdaptation metric selection

Bit Error Rate (BER)Bit Error Rate (BER)

Cannot be calculated in real-timeCannot be calculated in real-time

Error Vector Magnitude (EVM)Error Vector Magnitude (EVM)

1

2 e

I

Q

Error in thereceived symbolV2

V1 e = V1 - V2

Can be calculated in real-timeCan be calculated in real-time

EVM vs. BER

Channel effects

Adaptation Metric: EVM vs BER Adaptation Metric: EVM vs BER

Hence EVM is chosen as the real-time adaptation metric. Hence EVM is chosen as the real-time adaptation metric.

• Multi-path effects• Interference• Path loss• Noise

• Multi-path effects• Interference• Path loss• Noise

VIZOR operation-Operation close toerror threshold

VIZOR operation-Operation close toerror threshold

Normal operationNormal operation

Save powerSave power

VIZOR Operation : ConstellationVIZOR Operation : Constellation

EVM=22%EVM=22%

EVM=8%EVM=8%

EVM=35%EVM=35%

EVM=14%EVM=14%

VIZOR operationVIZOR operation

Channel 1Channel 1

Channel 2Channel 2

Receiver hardware and Power SavingsReceiver hardware and Power Savings

Tuning Knob 2

Performance

Tuning knob 1

Minimum power, maximum EVM locus

Increasing performance P

Tuning knob: Vdd, Vbias of different RF block, ADC wordsize

Power vs. Performance locusPower vs. Performance locus

OutlineOutline

Environment Adaptability in Transceivers

- Adaptive RF receivers. - Adaptive RF transmitters.

Process Adaptability -Process Sensing Using Test.

-Tuning for process variation.

Experimental Results

VIZOR: TransmitterVIZOR: Transmitter

RF PA

Drain Bias

Gate Bias

Power Management

DAC

Mixer

Baseband DSP

PARReduction

Control Law

Channel QualityChannel Quality

Dynamic Adaptive

Companded Signal;

PAR reduction = 5.6 dB

OFDM: High Peak to Average Ratio (PAR). Makes Power Amplifier (PA) inefficient due to high back off required.

3 dB back off halves a PA efficiency.

Pin

Pout

PA characteristics

Original Signal

P1dB

5.6 dB

Companding and PAR reductionCompanding and PAR reduction

Compliance to BER threshold Compliance to BER threshold

PAR Reduction

BER moderate channel

BER good channel

EVM moderate channel

EVM good channel

µ : Companding Factor

PAR reduction under different channels PAR reduction under different channels

64 QAM BPSK

5% 14% 35%

16 QAM QPSK

*EVM=

+EVM threshold16 QAM =12.5 %

+EVM thresholdQPSK=33 %

Good ChannelEVM=5.15 %

Moderate ChannelEVM=8.5 %

6.3 dB

4 dB

Moderate ChannelEVM=24.4 %

7.25 dB

5.6 dB

Good ChannelEVM=15.5 %

*From SNR boundaries (IEEE) +From BER threshold

Transmitter Power SavingsTransmitter Power Savings

3X3XPdc static PA

OutlineOutline

Environment Adaptability in Transceivers

- Adaptive RF receivers. - Adaptive RF transmitters.

Process Adaptability -Process Sensing Using Test.

-Tuning for process variation.

Experimental Results

Minimum power, maximum EVM locus

Increasing performance P

PROCESS

Effect of process variations: run RF BIST to pick the “Right” locus under variation.

Effect of Process variationEffect of Process variation

Performance

Tuning Knob 2

Tuning knob 1

Process SensingProcess Sensing

Transmitter spec estimation

Receiver spec estimation

BASEBAND DSP

Rx Module

Embedded sensor

Loopback hardware

TxModule

Transmitter spec estimation

Receiver spec estimation

BASEBAND DSP

Rx Module

Embedded sensor

Loopback hardware

TxModule

*V. Natarajan et. al. “ACT: Adaptive Calibration test….” VTS 08*V. Natarajan et. al. “ACT: Adaptive Calibration test….” VTS 08

OutlineOutline

Environment Adaptability in Transceivers

- Adaptive RF receivers. - Adaptive RF transmitters.

Process Adaptability -Process Sensing Using Test.

-Tuning for process variation.

Experimental Results

Process tuningProcess tuning

Design PhaseDesign Phase Production Test/Tuning PhaseProduction Test/Tuning Phase

Minimum power & Maximum

EVM locus for Np instances

Process Sensing & Process Tuning

Process Sensing & Process Tuning

Locus corresponding to the BEST MATCH (LMS)

DUT

Np process instances

For each processinstance

Perform environmental Adaptation during run timePerform environmental Adaptation during run time

Process 1Process 2

Process Np

PerformanceGain, IIP3, P/ Vdd, P/ Vb

..

Gain, IIP3, P/ Vdd, P/ Vb

Process adaptation metric for Np instances

Performance

Measure TX process

Measure RX process

Find BEST MATCH with available Np process instances

Measure Power sensitivities

With tuning

Without tuningWithout tuning

Channel Index

ReceiverPower

(W)

Power Savings through process tuningPower Savings through process tuning

OutlineOutline

Environment Adaptability in Transceivers

- Adaptive RF receivers. - Adaptive RF transmitters.

Process Adaptability -Process Sensing Using Test.

-Tuning for process variation.

Experimental Results

VIZOR system VIZOR system

VIZOR TX and RX

Runtime experimental results: ReceiverRuntime experimental results: Receiver

EVM EVM increasingincreasing

OFDM signalOFDM signal

Power Power consumption consumption decreasingdecreasing

System in OperationSystem in Operation

Environment Adaptability allows transceiver operation with very less built in design margin. (i.e. Virtually Zero Margin) - Saves significant power under favorable environmental condition.

- 3X transmitter and 4X receiver power savings could be achieved.

Process Adaptability makes this adaptation near optimal even under process variation.

ConclusionConclusion

Significant increase in battery life.Significant increase in battery life.

Questions ? Questions ?

Environment Adaptability: OverviewEnvironment Adaptability: Overview

BasebandReceiver

Transmitter

BasebandTransmitter

Receiver

Base Station (BS)Base Station (BS) Mobile Station (MS)Mobile Station (MS)

1 bit Encoded in MAC Header

0: EVM > EVMthreshold

1: EVM < EVMthreshold

Power saved in MS increasing battery life Power saved in MS

increasing battery life

Worst Case System Design

Worst Case System Design

Adaptive System Design

Adaptive System Design

Varyingchannel conditions

VIZOR OptimizerVIZOR Optimizer

Zero-margin operation – Save more power under favorable conditions (good channel)!!

Identify tunable Parameters

• LNA supply• LNA bias

• Mixer supply• Mixer bias

• ADC word sizeSet EVM threshold for satisfactory operation

Generate different Channels

• Interference• Multi-path

•Noise

Optimization

Optimal values oftunable parameters for

• Different channelconditions

• Different modulations (data rates)

PAR reduction and compliance to BER threshold PAR reduction and compliance to BER threshold

• Estimate transmitter specs• Tune the transmitter – Hardware + software knobs

-Process variation, non-idealities• Estimate Receiver specs

Transmitter spec estimation

Receiver spec estimation

BASEBAND DSP

Rx Module

Embedded sensor

Loopback hardware

Tuning

control

Tx Module

Tu

ned

ou

tpu

t

Baseband response

Process Sensing: Adaptive Calibration TestProcess Sensing: Adaptive Calibration Test

• Control law: LS error between golden and actual• Hardware knobs: Power supply, bias control• Software knobs: Reverse distortion polynomials

Control law

Test input

System under consideration

Knob Controll

er

Software knobs Tuned

output

Actual responseGolden

Hardware knob control

Envelope Detector

Appendix: ACT: Tuning techniqueAppendix: ACT: Tuning technique

Transmitter parameter estimation

Receiver parameter estimation without transmitter tuning

Receiver parameter estimation after transmitter tuning

Process estimation accuracyProcess estimation accuracy