Determining a low-pass filter combination for the CADDMAS integrated sensor

Author

Phillip D. Howell

Date of Award

5-1993

Degree Type

Thesis

Degree Name

Master of Science

Major

Electrical Engineering

Major Professor

Bruce W. Bomar

Committee Members

Gary Flandro, Roy Joseph

Abstract

This thesis determines an analog and digital filter combination for the Computer Aided Dynamic Data Monitoring and Analysis System (CADDMAS) integrated sensor. This filter combination processes analog instrumentation signals affected by broadband white and impulsive noise. The filter combination must: a) low-pass filter the analog signal to prevent aliasing, b) correct for magnitude and phase distortions introduced by the analog filter, c) produce two digital sample rates: f_ss, and d) band limit to f_s/2 both sample rates for maximum noise rejection.

These requirements can be met by a series combination of an analog anti-aliasing filter, a digital correction filter, a digital decimation filter (which produces f_s), and a series of digital interpolation filters (which produces 8f_s.). The anti-aliasing filter provides low-pass filtering of the signal in the analog passband (0 to f_s/2) with maximum attenuation for aliasing prevention above 1.5f_s

Four types of analog filters are considered. The final selection of an appropriate analog filter depends upon two factors. The first is the filter's impulse response. This is important due to expected ringing in the presence of broadband impulsive noise. The second factor is the ability of the filter to provide acceptable performance at the lowest possible analog sample rate (2f_s). It is found that the Butterworth, inverse Chebyshev, and elliptic filters meet these criteria.

After the analog filter output is converted to digital form, a digital filter corrects the magnitude and phase distortions introduced by the analog filter. It is found that each correction filter can correct for errors in the matching analog filter with the same digital filter length. The only difference among the correction filters is the number of impulse response coefficients with magnitude greater than unity.

The filter combination must produce two digital sample rates. The first, at f_s, is used for frequency domain analysis. This sample rate is achieved by decimating the signal by a factor of 2. In addition, the transition between decimation and interpolation stages is minimized in terms of the combined computational loading by adjusting the band edges of each stage.

The second sample rate (8f_s), used for time domain analysis, is achieved by increasing the sample rate by a factor of 8 through digital interpolation. The interpolation stage is presented as two options. The first is a series of three half-band interpolation filters. The second is a series combination of one half-band and one fourth-band interpolation filter. It is determined that both options have excellent response characteristics, however, the second option affords a 200 nanosecond savings in terms of computational loading.

Though it is determined that five filter combinations achieve acceptable performance, two combinations are singled out for special notice. These combinations have been chosen due to the fact that each meet the design specifications with the lowest analog filter order. These combinations are: the 4th order inverse Chebyshev and 4th order elliptic filters in series with a digital correction filter and a decimate-by-2 FIR filter followed by either of the two interpolation filter options.

Recommended Citation

Howell, Phillip D., "Determining a low-pass filter combination for the CADDMAS integrated sensor. " Master's Thesis, University of Tennessee, 1993.
https://trace.tennessee.edu/utk_gradthes/11911