Self-Testing and Self-Tuning Mixed- Signal/RF: Test and Validation

Transcription

1 Self-Testing and Self-Tuning Mixed- Signal/RF: Test and Validation A. Chatterjee Reliable Mixed-Signal Computing Group Georgia Institute of Technology USA

2 Overview Future real-time Computing/DSP/Mixed-signal/RF systems: the need for self-awareness Self aware: Environment and Health Built-In Test, Control and Adaptation

3 Remote Wireless Camera System Environment Quality of image/video Channel quality Health of the electronics (sensor + processor + RF) Control Power consumption vs accuracy of the image sensor + processor Power consumption vs accuracy of the RF front end

4 Ex 2. Autonomous Vehicles Source: General Motors

5 Key Design Paradigm The manner in which a real-time system reacts to external stimulus depends on its operating environment (workload) and its health Need for flexible algorithms flexible hardware flexible circuitry That can be dynamically tuned for accuracy vs. power vs. reliability

6 Adaptive Wireless Systems Recalibration Channel/Signal Quality Health Adaptation Control: Hardware/Software Sensor Bandpass Filter Low Noise Amplifier Mixer VGA Lowpass Filter M Phase detector PLL N Crystal Oscillator VCO Lowpass Filter Buffer Switch Control ADC DSP ADC Detection mechanism: EVM, Null tone, Fading System: Mixer Lowpass Filter DAC RF Front End Bandpass Filter Power Amplifier Baseband Amplifier Processor

7 Health Determination: BIST Tx parameters are computed from Envelope detector output Rx parameters are calculated from I and Q output while Tx parameters are known Can be extended to full loopback testing without the envelope detector

8 Time-Zero Tuning Analysis and control engine (ACE)

9 Summary: What is needed Design of signal quality and system health sensors hardware, software Built-in testing of all critical specs in minimal time Design of tunable RF/mixed-signal components tunable mixer, LNA, PA, VCO, data converter designs bias, supply voltage, tunable passives, clock Design of on-chip hardware for implementing signal quality + system health based real-time tunable component control no RF test instrumentation! Perform integrated cross-layer optimization and control!

10 E.g.:Tunable LNA

11 E.g.:Tunable LNA

12 BIST for Multiple Specs SIGNATURE TEST! Production Phase Multi-tone DUT with process input variations Test input Capture Measurement Space DSP M A R S Gain i IIP3 i Spec i N Mapping function

13 Signature test driven validation and tuning Multi-tone DUT with process input variations Test input Capture Measurement Space DSP M A R S Gain i IIP3 i Spec i N Mapping function MODEL VALIDATION / BIST NONLINEAR SOLVER MODEL

14 Signature-BIST: Overview Test Stimulus RF to lowfrequency conversion Test Response Diagnosis Two different techniques for tuning

15 Signature Based Diagnosis Using Regression Mappings Test response Best initial estimate of filter calibration coefficients Behavioral parameter estimation 1 For adaptive tuning For lookup-table 2 based correction

16 Signature Driven Model Parameter Estimation: TX + RX I-Q mismatch Gain and IIP3 specs Gain IIP2 IIP3 IIP5 AM-AM distprtion AM-PM distortion I-Q mismatch VCO phase noise DC offset NF

17 Signature Test Stimulus Designed so that model parameters can be uniquely determined from DUT response

18 Model Estimation OBSERVED RESPONSE MODEL LEAST SQUARES ERROR NONLINEAR SOLVER

19 Signature Based Model Parameter Estimation

20 Measuring Amplitude and Phase Distortion

21 Test Setup

22 Gain and Phase Distortion

23 Validation Driven Tuning

24 Example: Adaptive LNA TSMC.18um CMOS Design

25 Experimental Results: BIST I/Q imbalance Transmitter Specs Receiver Specs

26 Adaptation Performance: LNA

27 Adaptation Performance: LNA

28 Experimental Results: Transmitter Large parameter instances Small parameter instances

29 Experimental Results: Nominal Specs Gain IIP2 IIP3 Nominal 42.5 db dbm Lower bound 41.5 db dbm Upper bound 43.5 db dbm One-Instance (P1) Gain IIP2 IIP3 Before dBm - 8dBm -6 dbm After possible knob combinations (P1) for yield recovery Power conscious knob combination (P1) : W Converged Knob combination (P1) : W

30 Sensing Noise

31 Jitter Expansion from At-speed to Base-band At-speed Signal w/ Phase Noise Envelope Function: Env(t) Combined Signal: Y(t)

32 Simulation (Jitter Histogram in Random Phase Noise) Jitter Increased by 20 (20 = 1GHz / 50MHz) 0.01 ns 0.2 ns At Real Speed: 1 GHz At Base-band: 50 MHz The Same Phase Noise:

33 Simulation (Jitter Histogram in Random + Sinusoidal Phase Noise) ns ns At Real Speed: 1 GHz At Base-band: 50 MHz Jitter Increased by 20 (20 = 1GHz / 50MHz)

34 Hardware Implementation (Voltage Combiner + Envelope Detector + Comparator)

35 Measurement II Phase Noise Plot: PLL Board Output (1) Noiseless Reference (2.401 GHz) (2) Noisy Source (At-speed: GHz) (3) Envelope Detector Output (Base-band: 1 MHz) oisy Source: Analog Device Eval-ADF4360

36 Measurement Deterministic Phase Noise ( 50kHz )

37 PRBS Jitter PRBSequence Test architecture

38 PRBS Jitter 2 Gbps MHz reference 1 Mhz sensor O/P

39 PRBS Jitter 2 Ghz clock without and with 8.5 ps rms jitter (50 ps/div) Envelope detector without and with injected jitter (50 ns/div)

40 PRBS Jitter 2 Gbps K28.5 pattern without and with 17 ps rms jitter Envelope detector output without and with injected jitter

41 Data Acquisition Via Incoherent Undersampling

42 Data Acquisition Via Incoherent Undersampling Direct single shot acquisition Reconstructed waveform from single shot data

43 Real-time adaptation: Save power when channel/signal quality is good DSP computes QoS metric for receiver: Receiver Adapts Tower computes QoS metric for transmitter and sends back to receiver: Transmitter Adapts

44 Overview: Extend the channel into the transceiver algorithm architecture RF/mixedsignal sensing Tuning knobs Software + hardware sensors produce signatures Vertically integrated control for adaptation Sense: Subsystem health, Channel, Signal quality control Sensing: Channel quality + Quality of the RF front end Integrated Adaptation and Test Algorithms Control: Transmitter/Receiver performance vs. power modulation via tuning knob control Adaptive Low Power Circuits for Wireless Communications, A. Tasic, W. A. Serdijn and J. R. Long, Springer ISBN , Energy-scalable OFDM transmitter design and control, B. Debaillie, et. al.., 43 rd DAC, July 2006, pp (IMEC, Belgium)

45 Channel Sensing: Checker Computes metric which measures channel quality + transceiver fidelity Transceiver fidelity = f( control algorithms) Candidates Channel estimation algorithms (IMEC, Belgium) EVM + SNR + RSSI + frame drop rate (Hitachi) EVM + RSSI + No of bits corrected Can be computed in tens of ms EVM = f(gain, IIP3, I/Q mismatch, phase noise) + f(interference, multi-path, fading, channel noise)

46 Low Power Operation EVM = 22% EVM = 38% Normal operation VIZOR operation -Operation close to error threshold Save power EVM = 8% EVM = 13.5%

47 Key Issue As channel/signal quality is getting better or worse, how do we simultaneously control tuning knobs (10 20 parameters) while minimizing power and not exceeding maximum allowed BER? LNA Mixer VCO Digital Predistortion PA DAC ADC

48 VIZOR: 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 size Generate different Channels Interference Multi-path Noise Optimization Set EVM threshold for satisfactory operation Optimal values of tunable parameters for Different channel conditions Different modulations (data rates)

49 OPTIMIZER RF parameters, Vdd, Bias Increasing performance P Worse Channel Decreasing performance P Better Channel Minimum power, maximum EVM locus Wordsize Distortion

50 VIZOR: Receiver

51 VIZOR: Transmitter 3X power savings!

52 Optimal Values for Different Channels RF front-end (LNA and Mixer) operation Supply, bias 0 = best channel. 1 = worst channel Power

53 Receiver Power Savings and hardware

54 X-Treme RF: Process + Channel Process Tuning Feedback control = f(tuned process) Feedback Optimizer Hot Checker

55 OPTIMIZER RF parameters, Vdd, Bias Increasing performance P PROCESS Minimum power, maximum EVM locus Wordsize Distortion EFFECT OF PROCESS VARIATIONS: RUN RF BIST TO PICK THE RIGHT LOCUS!

56 Need for power tuning Power tuning of the receiver enables operation at lower power consumption levels for majority of the channel conditions!!

57 Conclusions: Technology Enablers Design of signal quality and system health sensors hardware, software Design of tunable RF/mixed-signal components tunable mixer, LNA, PA, VCO, data converter designs bias, supply voltage, tunable passives, clock Methodology + software to interpret system health from sensor data Design of on-chip hardware for implementing signal quality + system health based real-time tunable component control Perform integrated cross-layer optimization and control!