Browse Prior Art Database

Using FFTs to Filter Multi-input Analog Time Domain Signals Real Time

IP.com Disclosure Number: IPCOM000121365D
Original Publication Date: 1991-Aug-01
Included in the Prior Art Database: 2005-Apr-03
Document File: 4 page(s) / 120K

Publishing Venue

IBM

Related People

Young, DH: AUTHOR

Abstract

This article presents a concept for digitally filtering multiple inputs using a FFT in the time domain. This concept can be implemented in minimal hardware versus an analog filter or classic digital filter approach in a multi-channel environment. Current 1.0 um VHSIC technology has allowed an evolution of dedicated FFT hardware with the processing power necessary to perform the following multi-channel filter approach.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Using FFTs to Filter Multi-input Analog Time Domain Signals Real Time

      This article presents a concept for digitally filtering
multiple inputs using a FFT in the time domain.  This concept can be
implemented in minimal hardware versus an analog filter or classic
digital filter approach in a multi-channel environment.  Current 1.0
um VHSIC technology has allowed an evolution of dedicated FFT
hardware with the processing power necessary to perform the following
multi-channel filter approach.

      Basic analog low-pass filters must still be used prior to
digitizing the input signal; therefore, analog filters cannot be
totally eliminated.

      This multi-channel filter can perform the necessary
bandlimiting on a digitized data stream.  This data stream can
consist of a group of time division multiplexed channels fed
sequentially into the FFT and down sampled in the time domain at the
output of the FFT.  These bandlimited time domain samples can then be
presented to a second FFT to acquire the spectral content of the
signal.

      An FFT is basically a bank of sample and integrate filters.
The filter parameters of the filter are a function of the bin
resolution of the FFT.  There are some secondary effects due to the
window function before the FFT.  To achieve the time domain sampling,
the FFT input data  stream must contain data from the current time
sample as well as previous time samples out of the analog front end.
This is the technique of shifted or hopped FFTs in time.  Refer to
the figure for a graphical representation of this.

      The number of new time points processed is referred to as the
skip interval (SI).  This interval establishes what the down sampled
rate of the multi-channel filter is.  This new sampling rate
determines the resolution of the second FFT.

      The following is a sample calculation to establish the FFT size
and SI based on a filter bandwidth requirement of 240-720 Hz.  The 3
dB points of the filter are determined by the 3 dB bandwidth of the
selected window prior to the FFT. For this example, a minimum three-
term Blackman-Harris window is used.  Its 3 dB bandwidth is 1.66 bins
(*).

      The sampling rate and FFT size are selected as follows.
      (1) fs=10kHz  N1=16

      The FFT size is selected to give a bin resolution that
encompasses the bandwidth of the filter being implemented.

                            (Image Omitted)

      Bin 1 is the bin that will be time sampled out of the FFT.
If the filter bandwidth had been higher, a different bin could be
sampled or a different sized FFT could be used. The hardware
implementation used a Radix-4 engine in the FFT; therefore, the FFT
size had to be powers of four.

      The 3 dB points of the filter are a function of the minimum
three- term Blackman-Harris window.

                            (Image Omitted)

      Subs...