User`s manual

Manual revision 016 Appendix V: Real-Time Data Correction Algorithms SBE 49
68
Filtering Data
Filtering smoothes the data, to match the temperature and conductivity sensor
time constants. In post-processing, Sea-Bird recommends using a low-pass
filter (Filter module in SBE Data Processing) on recorded FastCAT data, with
a time constant of 0.085 seconds for both temperature and conductivity. This
filter algorithm cannot be implemented for real-time data processing in the
FastCAT, because the low-pass filter is run forward through the data (an
impossible task when processing in real-time) and then run backward through
the data. Instead, a 5-scan cosine window filter (based on the Window Filter
module in SBE Data Processing) was implemented for real-time data
processing in the FastCAT; this window filter provides results comparable to
the low-pass filter.
The window filter calculates a weighted average of data values about a center
point, using the following transfer function:
L/2
y(n) = Σ w(k) x(n-k)
k=-L/2
It then replaces the data value at the center point with this average.
The window filtering process follows:
1. Calculate cosine filter weights (see the equations below).
w(n) = 1 for n = 0
w(n) = cos for n = - . .-1, 1 . .
2. Normalize filter weights to sum to 1.
Applying Conductivity Cell Thermal Mass Correction to Data
The FastCAT uses a recursive filter to remove conductivity cell thermal mass
effects from the measured conductivity. Recommended values for alpha and
tau (tau = 1 / beta) for the FastCAT are 0.03 and 7.0 respectively.
The algorithm, based on the Cell Thermal Mass module in SBE Data
Processing, is:
a = 2 * alpha / (sample interval * beta + 2)
b = 1 - (2 * a / alpha)
dc/dT = 0.1 * (1 + 0.006 * [temperature - 20])
dT = temperature - previous temperature
ctm [S/m] = -1.0 * b * previous ctm + a * (dc/dT) * dT
corrected conductivity [S/m] = c + ctm
Note:
In the window filter equations:
• L = window length in scans
= 5 for FastCAT real-time data
processing algorithm
• n = window index,
-( [L-1] / 2) to +( [L-1] / 2),
with 0 the center point of the window
•
w(n) = set of window weights
n x π
L + 1
L - 1
2
L - 1
2