slides electrical instrumentation lecture 7 version 2

1

ET8.017El. Instr.Delft University of Technology

1

Electronic Instrumentation

Lecturer: Kofi Makinwa

[email protected]

015-27 86466Room. HB13.270 EWI building

Teaching Assistant: Saleh Heidary

[email protected]


2

Introduction

� Goal of a measurement system:– Get the most out of a measurement or sensor– in terms of information quality

� How?– Filtering (averaging) and shielding– Compensation (e.g. feed-forward)– Correction (e.g. auto-zeroing)– Feedback– Modulation and correlation

2


34-2

Precise butnot accurate

Not accurate andnot precise

Accurate butnot precise

Accurate andprecise

Time TimeTimeTime

Stable butnot accurate

Not stable andnot accurate

Accurate (on the average)

but not stable

Stable andaccurate

0

f fff

Accuracy, precision (resolution) and stability


4

Signal processing techniques

Add-on techniques:� filtering

– prefiltering

– postfiltering

– correlation

� correction– post-processing

(e.g. auto-zeroing, linearization)

Built in techniques:� compensation

– balancing– bridge

� feedback– requires an inverse

transducer (actuator)

� modulation– chopping– spinning– swapping

3


5

Filtering: general principle

� reduces additive errors� in a specified frequency band� reduces the effects of thermal noise

interference

prefilter postfilter

sensor processormeasurand out


6

Pre-filtering

� Pre-filtering (prior to transduction)� Mitigates the effects of:

– electrically induced signals (capacitive) ⇒ shielding– magnetically induced signals (inductive)

⇒ reducing loop area; shielding– changes in environmental temperature ⇒ isolation– unwanted optical input ⇒ optical filters– mechanical disturbances (shocks, vibrations) ⇒ dampers– .....

4


7

Electrically-induced interference

, ,s

n o n ig

CV V

C=

sensor system interface system

Vn,i

Cs

Cg

Vn,o

s gC C<<

Vn,i: voltage noise source

Cs: capacitance between noise source and signal leads

Cg: capacitance between signal leads and ground


8

Shielding

, , 0restn o n i

sh

CV V

C= →

Vn,i

Vn,o

Cg

Csh

Crest

interface systemsensor system

Crest: rest capacitance between noise source and signal leads

Cg: capacitance between noise source and ground

Csh: capacitance between shield and signal leads

5


9

Post-filtering

� post-filtering (after transduction)� in the electrical domain

– low pass, high pass, band pass– first order, .... higher order– characteristic (Butterworth, Bessel,...)– analog, digital– Matched filtering

� If the time-domain characteristics of the measurand are known, correlation and windowing techniques can be used


10

Post-filtering

� High-pass filter removes 50Hz interferencefrom ambient light sources

6


11

Correction

� Add-on to an existing system– model-based correction (after calibration)– correction of cross sensitivities (extra sensors)

� Correcting for additive errors– auto-zeroing– CDS

� Correcting for multiplicative errors– linearization– 3-signal method (calibration of linear systems)


12

Correction - Overview

interference

sensorsystem

processormeasurand out

modeldata

additional sensor

measureddata

reference

� Can cancel both additive and multiplicative errors� But additional sensors and multiple errors limit the

achievable accuracy

7


13

Compensation

� where?– pre-compensation (inherent design)– post-compensation (model-based signal correction)

� which strategy?– balancing– feedforward– feedback (with an actuator)


14

Balancing: General principle

� reduces (common) additive errors� improves dynamic range

Examples� dummy sensor: used to compensate for

environmental influences� Extra temperature sensor: used to compensate

for unwanted temperature dependencies

compensatingelement


interference

8


15

Bridge compensator

VRR

RRR

RV io

+−

+=

43

4

21

2

Vi

R1

R2

R3

R4

Vo 4132 RRRR =

( )3412 RRRR =?

• for accurate absolute measurements of resistances

• Insensitive to CM disturbances e.g. temperature

• but requires an accurate, adjustable resistor


16

Switched Cap “Resistor”

� Switched cap network emulates an adjustable resistance- Effective resistance value: Rref = Tclk/C2

M. Perrot et al., ISSCC 2012

9


17

Bridge indicator

VRR

RRR

RV io

+−

+=

43

4

21

2

Vi

R1

R2

R3

R4

Vo

• but output is sensitive to drift and error in Vi

initially: resistances have values such that

04010302 RRRR =

• measures small changes with respect to "an initial” value


18

Ratiometric readout

ADC and bridge use the same reference voltage Vref

⇒ errors in Vref do not affect digital output

10


19

Assignment 8

� Consider the ratiometric system shown on the previous slide. If the noise densities shown for the bridge, preamp and ADC, correspond to the situation when the bridge is balanced:

1. Calculate the resistance of each arm of the bridge2. If the measurement BW is 10Hz, how many bits of

resolution must the ADC have in order to ensure that the resolution of the system is determined by thermal noise rather than by quantization errors.

3. If the ADC has 20-bit resolution (not necessarily the answer to question 2 above) what percentage change in the resistance of a bridge element corresponds to a 1LSB change in its digital output.


20

Differential measurements

1 1 1

2 2 2

m m n n

m m n n

y S x S x

y S x S x

= += − +

( ) ( )1 2 1 2comp m m m n n ny S S x S S x= + + −

2 22

ncomp m m n n m m n

m

Sy S x S x S x x

S

∆= + ∆ = +

2 m

n

SCMRR

S=

∆

opposite sensitivity

by design!

nx

compensatingsensor

sensor

processor

measurand

out

interference

1y

mx

2y

compy

Sm1: sensitivity sensor 1 to measurand

Sm2: sensitivity sensor 2 to measurand

Sn1: sensitivity sensor 1 to interference

Sn2: sensitivity sensor 2 to interference

11


21

Differential capacitive displacement sensor

differential ∆x

+Vi -Vi

IoV1

I

V2

C2

C1

Cf

2

1 2o io f

CI j V

C Cω ∆= ⋅ ⋅

+

C2C1 Cf


22

Differential sensor interface

R

Vo

Cf

_

+

Vi

C2

-Vi

C1

2 io

f

VV C

C= ∆

1 2i i

of f

V VV C C

C C= − +

1

2

o

o

C C C

C C C

= + ∆= − ∆

� The resistance R reduces drift� But it introduces extra noise and gives rise to a

frequency-dependent transfer function

12


23

The utility of feedback

� If A(f)•β >>1 ⇒ ACL ≅ 1/β.

� For moderate 1/β, op-amp� DC gain is large enough!

� β ⇒ Resistor/Capacitor ratios.� Resistors ⇒ 0.01%, 5 ppm/°C� Capacitors ⇒ 1%, 500 ppm/°C

� ⇒ ACL can be very accurately defined.


24

Feedback: General principle

� reduces multiplicative errors� improves dynamic behaviour

interference

feedback


13


25

General application of feedback

� Prerequisites for effective error reduction are:– high forward path transfer– stable feedback path transfer

0

01f

SS

S k=

+ ⋅

0

0 0

1

1f

f

dS dS

S k S S= ⋅

+ ⋅

S0

k

yoxm

+

−−−−


26

Example 1: Accelerometer

sensor interface

seismic mass

differential capacitor

V(∆C)

Vo

actuator

∆x

a

inertia Vospring sensor interface gain

actuatorFa

Fi

a

∆x ∆C VC

+ −

Fs

14


27

Analysis of a feedback accelerometer (1)

1s e

o ia s e

H HV H a

H H H=

+

io

a

HV a

H=

inertia Vospring sensor interface gain

actuatorFa

Fi

a

∆x ∆C VC

+ −

Fs

Hi: transfer from acceleration a to inertial force Fi

Hs: transfer from force Fs on sensor to displacement ∆x

He: transfer from displacement to output voltage Vo

Ha: transfer of actuator: from output voltage Vo to Fa

: 1:a sif H H >>

Independent of system parameters of the spring, sensor, interface, gain factor

Transfer function depends only on mass and actuator


28

Example 2: Wind sensor

15


29

δT

� Airflow induces an on-chip temperature gradient δT� This is sensed by an on-chip thermopile and

cancelled by a pair of heaters� Transfer function only depends on the heaters

(resistors) and their supply voltage

Chip

Airflow

Heater

Thermal balancing

Controller

Heater

SensorCross-Section Thermopile


30

Wind sensor chip

� On-chip heaters

� Thermopiles ⇒⇒⇒⇒ wind induced temperature differences δTNS & δTEW

� On-chip controllers nulls δTNS & δTEW

� |δP| ⇒⇒⇒⇒ wind speed� arg(δP) ⇒⇒⇒⇒ wind direction

16


31

Smart wind sensor chip

� Digital outputs δPNS & δPEW

� After calibration:Speed error: ± 4% Angle error: ± 2°

� Comparable to conventional wind sensors

� But no moving parts!


32

Modulation: General principle

� Basic ideas: shift information to a quiet frequency bands OR make information more robust

� Reduces additive errors� In particular, offset, drift, LF noise and gain error

interference


modulation demodulation

17


33

Modulation types

� AM modulation: Chopping, synchronous detection, dynamic element matching

� FM modulation: Used to make information more robust to amplitude variations

� Pulse-width (or duty-cycle) modulation: also used to make information more robust to amplitude variations, also used for simple digital-to-analog conversion

modulator

carrier

in out


34

The Thermal Management Challenge!

� SoCs and multi-core processors ⇒⇒⇒⇒ multiple hot spots � Hot spot location depends on (dynamic) workload ⇒⇒⇒⇒ multiple (20+) temperature sensors

Intel 45nm Xeon Courtesy of S. Rusu Intel

18


35

Thermal Diffusivity Sensing

� Heater generates small heat pulses (mWs)� Temperature sensor picks up delayed pulses (mVs)� Resulting thermal delay is temperature-dependent � Exploits the strengths of CMOS technology:

– Lithography � accurate spacing s– Pure silicon substrate � DSi is well-defined– Fast circuits � accurate delay

measurements


36

A CMOS ETF

• Heater = n+-diffusion resistor

• Temp. sensor= p+/Al thermopile

• Can be made in any IC process!

• Circular, differential layout for maximum output

S

Heater Thermopile

100 µm

19


37

Smart TD Sensor

� At room temp, s =20µm, f = 100kHz ⇒⇒⇒⇒ φETF ~ 90°� Quartz-crystal clock generates fdrive (typ. 100s of kHz)

� φETF is digitized by a phase-domain Σ∆ modulator

� Resolution is mainly limited by ETF’s thermal noise

[van Vroonhoven, Makinwa, ISSCC 2008]

φ ∝ /drive SiETF s f D


38

Phase Digitizer

� How to accurately measure the phase of a small (1mVpp) and noisy signal?

� Use feedback, an ADC and a digital phase generator

� Then digitally filter the result (few Hz BW)

20


39

Phase-Domain ∆Σ∆Σ∆Σ∆ΣM

� Since the sensor’s BW is so low, we can over-sample its output and then average ⇒ a 1-bit ADC is enough!

� This technique is known as sigma-delta modulation

� 12-bit resolution in 0.5Hz BW � fs > 2kHz � easy!


40

Subtracting phase?

� Use a synchronous detector!

� DAC references are just phase-shifted copies of fdrive� multiplier inputs are at same frequency

� DC ∝ cos(φETF – φfb ) = 0 when φfb = φETF – 90°

� Approx. linear close to (φETF – φfb ) = 90°

21


41

Performance

– ±0.2ºC (3s) in 0.18µm CMOS (same ETF layout)

� Accurate, and scales!

� Untrimmed spread is mainly limited by lithography

– ±0.6ºC (3σ) in 0.7µm CMOS


42

Overview: Signal processing techniques

interference


basic system

compensatingsensor

feedback

modulation demodulation

measureddata

inherent design

prefilter postfilter

correction

modeldata

additional systems

22


43

THE END!


44

Assignment 9

� The thermal diffusivity of silicon is proportional to 1/T1.8 where T is absolute temperature

� If an electrothermal filter has a phase shift of 90° at 300K, calculate its phase-shift at the extremes of the military temperature range, i.e.-55°C and 125°C

� A smart TD sensor employs a phase digitizer. How many bits of resolution should it have in order to achieve a temperature-sensing resolution of 0.1°C?

� If the phase digitizer employs the feedback architecture shown on slide 38 and uses a chopper as a phase detector, what range of phases must its phase DAC provide in order for the sensor to cover the military range?

slides electrical instrumentation lecture 7 version 2

Documents

induced et8

precision et8

precise butnot accurate

accurate andnot accurate

filtering averaging

effects of thermal noise5

precise preciseffff

addon techniques