Sie sind auf Seite 1von 23

Spring 2017 CHEMISTRY 105

Instrumental Methods in Analytical Chemistry


Lecture: MW 9 AM 10 AM 219 Dwinelle
Labs: MTuWTh 1 5 PM 305 Latimer

Lecture 2. Error Analysis & Noise

Prof. Ke Xu

1
Normal (Gaussian)
distribution:

1 1 x 2
f ( x) = exp
2 2

Highest probability for close to


Wider spreading for larger

3
Confidence intervals (CI) for a single measurement, known

For a confidence level of 68%,


should be in the interval:
CI for = x z
x

Confidence
z
level
0.8 1.2816
For a confidence level of 95%,
should be in the interval: 0.9 1.6449
0.95 1.9600
1.96 x 1.96 0.98 2.3263
Most often used! 0.99 2.5758
0.995 2.8070
0.998 3.0902
0.999 3.2905
For a confidence level of 99.7%,
should be in the interval:
0.9999 3.8906
0.99999 4.4172
x 3 0.999999 4.8916

4
Propagation of uncertainty
N
= lim (x )
2
i N
For addition or subtraction of data N i =1

For x = x1 x2 x3 ... = 12 + 2 2 + 3 2 ...


Example: total weight

For multiplication or division of data

dxy ydx xdy dxy dx dy


dxy = ydx + xdy = + = +
xy xy xy xy x y

2 2
xy y
= x +
xy x y

Use relative uncertainly for the calculation!

2 2 2
x p q r
For x = p q / r... = + + ...
x p q r
Example: R = U/I
Everything is unit-less 5
6
Repeated measurements averaging of results
Repeated measurements of the same signal: each measurement has the same

Sum: Sum = x1 + x2 + x3 + ... + x N

2 2 2
Uncertainty of sum: sum = 1 + 2 + 3 ... = N 2 = N

1.0

Uncertainty of mean
Mean: x = Sum / N 0.8

0.6

Uncertainty of mean: m = sum / N = / N 0.4

0.2

Uncertainty of Uncertainty of 0.0


the averaged each single 1 2 3 4 5 6 7 8 9 1011121314151617181920
# of measurements
5
value measurement

Improvement in S/N
4

Signal Signal
S/N: = N 2

m 1

0
Improved S/N for repeated measurement! 1 2 3 4 5 6 7 8 9 1011121314151617181920
# of measurements
Improving S/N through averaging

5
Improvement in S/N
4

0
1 2 3 4 5 6 7 8 9 1011121314151617181920
# of measurements

7
Confidence intervals (CI) for mean, known

Confidence
z
level
0.8 1.2816
0.9 1.6449
0.95 1.9600
0.98 2.3263
0.99 2.5758
0.995 2.8070
0.998 3.0902
0.999 3.2905
m = / N 0.9999 3.8906
0.99999 4.4172
For a confidence level of 95%, should be in the interval: 0.999999 4.8916

x 1.96 m
CI for = xz
N

x 1.96
N
8
9
CI when is unknown: using s to estimate
With known With unknown

Normal distribution: Student's t-distribution


X-axis is in unit of X-axis is in unit of s

Degree of freedom: = N 1
Approaches normal distribution for
Distribution broadens for smaller : extra
uncertainty when using s to estimate
CI when is unknown
Values of t for different confidence levels

80% 90% 95% 98% 99% 99.9%


1 3.078 6.314 12.710 31.820 63.660 636.600
2 1.886 2.920 4.303 6.965 9.925 31.600
3 1.638 2.353 3.182 4.541 5.841 12.920
4 1.533 2.132 2.776 3.747 4.604 8.610
5 1.476 2.015 2.571 3.365 4.032 6.869
6 1.440 1.943 2.447 3.143 3.707 5.959
7 1.415 1.895 2.365 2.998 3.499 5.408
s
CI for = xt 8 1.397 1.860 2.306 2.896 3.355 5.041
N
9 1.383 1.833 2.262 2.821 3.250 4.781
10 1.372 1.812 2.228 2.764 3.169 4.587
Compared to known :
15 1.341 1.753 2.131 2.602 2.947 4.073

= xz 20 1.325 1.725 2.086 2.528 2.845 3.850
N
50 1.299 1.676 2.009 2.403 2.678 3.496
t>z
100 1.290 1.660 1.984 2.364 2.626 3.390
t z for
1.282 1.645 1.960 2.326 2.576 3.291
10
Example (a1-6)
95% CI for three measurements: 0.084, 0.089, and 0.079
N

x i
0.084 + 0.089 + 0.079
x= i =1
= = 0.084
N 3

(x x )
N
0 + (0.005) 2 + (0.005) 2
(N 1) =
2
s= i = 0.005
i =1 2

s
95% CI: = x t = 2, for 95% CI: t = 4.303
N
0.005
= 0.084 4.303 = 0.084 0.012
3

However, if is known to be 0.005 from previous tests:

0.005
95% CI: = x z = 0.084 1.96 = 0.084 0.006
N 3 11
Experimental sources of noise/uncertainty

Improve S/N by reducing noise!

Chemical Noise
Undesired chemical reactions / interactions
Interfering chemical compounds: needle in a haystack
Fluctuation in humidity

Reduction:
Improve selectivity
Separation
Multidimensional analysis

Instrumental Noise
Associated with each component of the instrument

Thermal (Johnson) Noise


Shot Noise
Flicker (1/f) Noise
Environmental Noise
12
Instrumental noise white noises (f-independent)
Thermal (Johnson) Noise
Thermal (kT) agitation of electrons and other charge carriers

V rms = 4kTRf
Bandwidth
Temperature Resistance
Reduction:
Lower R of the circuits
Lower T: e.g., Liquid-nitrogen cooled detectors
Lower bandwidth of instrument
1
However, f = f t r
3t r Response time (rise time)

Shot Noise
When charged particle crosses a junction. Quantized events statistical fluctuation

= N irms = 2 Ief
Reduction:
Lower bandwidth of instrument 13
Instrumental noise 14

Flicker Noise (1/f noise; pink noise)


Higher noise at lower frequency!

Environmental noise
Hardware devices for noise reduction
Grounding and shielding

15
Hardware devices for noise reduction
Difference and instrumentation amplifiers
Analog filtering

16
Hardware devices for noise reduction

Modulation
Moves the measurement to higher f (reduces flicker noise)
Separation of real signal at the modulated frequency from random noise

Signal chopping

17
Lock-in amplifier

Picks up only signal at the exact same frequency as the reference signal
Reject signals/noises at other frequencies

18
Improving S/N through averaging -repeated experiments

5
Improvement in S/N
4

0
1 2 3 4 5 6 7 8 9 1011121314151617181920
# of measurements

19
20
For moving targets, averaging and smoothing

Moving
average

Works well for continuously evolving signal


Boxcar averaging Loose of signal details lower resolution in x
Digital filtering using Fourier transform

Cutoff frequency

21
Piece-wise fitting and smoothing

Smoother results vs. preservation of signal details


22
Homework due a week from today in class
Alice and Bob are re-enacting Cao Chongs weighing experiment. They found two
types of rocks, one type being ~50 kg in weight, and the other being ~20 kg in
weight. Now for an elephant that weighs ~1 ton, the measurement could be done
with 20 of 50 kg rocks, or 50 of 20 kg rocks.
Alice argues that the first approach gives a more precise result as more
rocks will introduce more error. Bob argues that the second approach is better as
the error from different rocks tends to cancel out each other.
Calculate the uncertainty of the final result for the two approaches for the
following two cases: (A) The absolute measurement uncertainty for each rock is
fixed to be 0.5 kg. (B) The relative measurement uncertainty for each rock is fixed at
2%. Which of the above two approaches win out for the two cases? (C) Is there a
mathematical form of error-weight relationship that would lead to equal uncertainty
of the final result for the two approaches?

+ Problem a1-12 (in Appendix 1)


& Problem 5-8 in text book
23
Thanks!

http://xkcd.com/1347/

Das könnte Ihnen auch gefallen