|
40m
QIL
Cryo_Lab
CTN
SUS_Lab
TCS_Lab
OMC_Lab
CRIME_Lab
FEA
ENG_Labs
OptContFac
Mariner
WBEEShop
|
| 40m Log |
Not logged in |
 |
|
|
|
Message ID: 394
Entry time: Sat Mar 22 22:39:02 2008
|
| Author: |
mevans |
| Type: |
Summary |
| Category: |
CDS |
| Subject: |
Direct Form 2 filters are bad |
|
|
Here I show a comparison between the filter algorithm currently used in LIGO (Direct Form II), and an alternative algorithm designed to reduce numerical noise. The input signal is
x = sin(2 * pi * t) + 1e-9 * sin(2 * pi * (fs / 4) * t);
where fs = 16384 is the sample rate. The filter is a 4th order notch at 1Hz (f_poles = f_zeros = 1Hz, Q_poles = 1, Q_zeros = 1e6). It is clear that the DF2 algorithm produces a noise floor that is, for this simple filter, 1e-11 / rtHz smaller than the input drive amplitude (see plots). That should probably be scary given how many second-order-sections we run our signals through. The low-noise form does a somewhat better job. The low-noise algorithm has the same memory and computational requirements as DF2, and our CDS guys have the code in hand. I suggest we start testing soon.
(The code is included below. You will need my Matlab library to run the top level test script.) |
|
|
|
|
| Attachment 3: |
FiltRT.zip
1 kB
Uploaded Mon Mar 24 14:07:26 2008
| |