Scribd
Upload
37 views5 pages

Least Square Filter

1) The document describes an example of least-squares filtering to estimate an input signal u(t) given noisy and quantized measurements y(t). u(t) is piecewise constant over intervals of 1 second from 0 to 10 seconds. y(t) is obtained by filtering u(t), sampling at 10Hz, and 3-bit quantizing the samples. 2) A linear model is constructed where y = Ax + v, with A relating the samples in y to the input levels in x, and v representing quantization error. 3) Applying least-squares estimation gives an estimate of x with lower RMS error than without filtering. Rows of the left inverse show how different measurements

Uploaded by

duc2711
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
37 views5 pages

Least Square Filter

1) The document describes an example of least-squares filtering to estimate an input signal u(t) given noisy and quantized measurements y(t). u(t) is piecewise constant over intervals of 1 second from 0 to 10 seconds. y(t) is obtained by filtering u(t), sampling at 10Hz, and 3-bit quantizing the samples. 2) A linear model is constructed where y = Ax + v, with A relating the samples in y to the input levels in x, and v representing quantization error. 3) Applying least-squares estimation gives an estimate of x with lower RMS error than without filtering. Rows of the left inverse show how different measurements

Uploaded by

duc2711
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

EE263 Autumn 2015 S. Boyd and S.

Lall

Example: Least-squares filtering

1
Example: estimation / filtering

u w y
H(s) A/D

I signal u is piecewise constant, period 1 sec, 0 ≤ t ≤ 10:

u(t) = xj , j − 1 ≤ t < j, j = 1, . . . , 10

I filtered by system with impulse response h(t):


Z t
w(t) = h(t − τ )u(τ ) dτ
0

I sample at 10Hz: ỹi = w(0.1i), i = 1, . . . , 100

I 3-bit quantization: yi = Q(ỹi ), i = 1, . . . , 100, where Q is 3-bit quantizer


characteristic
Q(a) = (1/4) (round(4a + 1/2) − 1/2)

I problem: estimate x ∈ R10 given y ∈ R100

2
Linear model

1.0 1.25

0.5 1.00
0.75
u(t) 0.0 s(t) 0.50
-0.5 0.25
-1.0 0.00
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10

1.0 1.0

0.5 0.5

w(t) 0.0 y(t) 0.0

-0.5 -0.5

-1.0 -1.0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10

we have y = Ax + v, where
Z j
I A ∈ R100×10 is given by Aij = h(0.1i − τ ) dτ
j−1

I v ∈ R100 is quantization error: vi = Q(ỹi ) − ỹi (so |vi | ≤ 0.125)

3
Results from least-squares estimation

1.0

0.5

0.0

-0.5

-1.0
0 1 2 3 4 5 6 7 8 9 10

I plot shows least-squares estimate: xls = (AT A)−1 AT y in red

kx − xls k
I RMS error is √ = 0.03
10
I better than if we had no filtering! (RMS error 0.07)

4
Rows of the left-inverse
0.15

0.10

row 2 0.05
0.00

-0.05
0 1 2 3 4 5 6 7 8 9 10

0.15

0.10

row 5 0.05
0.00

-0.05
0 1 2 3 4 5 6 7 8 9 10

0.15

0.10

row 8 0.05
0.00

-0.05
0 1 2 3 4 5 6 7 8 9 10

I some rows of Bls = (AT A)−1 AT

I rows show how sampled measurements of y are used to form estimate of xi


for i = 2, 5, 8
I to estimate x5 , which is the original input signal for 4 ≤ t < 5, we mostly use
y(t) for 3 ≤ t ≤ 7
5

You might also like