Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
g~ 34~3~
~l--
Low Frequency Digital Notch Filter
Technical Field
The present invention concerns improvements in or
relating to the design of low frequency digital notch
filters and, composite filters incorporating the same.
There are many Digital Signal Processing (DSP)
applications where one of the functions required is a
filter to reject any low frequency signals, and possibly
in addition, to reject any DC component of the signal. A
typical application of such a filter arises where the
digital signal being processed is a digital representation
of an analogue signal where the analogue signal contains
unwanted signals at 50Hz or 60Hz. These signals can arise
as signals coupled into the system from nearby line
powered equipment. It is often necessary to remove these
unwanted signals together with any DC component that is
present.
Background Art
A number of different filter structures have hitherto
been considered for implementing such a filter,
particularly for applications where the frequencies of
signals that are to be rejected are but a small fraction
of the system sampling rate (approx. 1%). When considering
the suitability of such structures, a number of aspects of
their performance must be considered. To ensure easy and
~ ~8~4~33
efficient implementation, it is desirable that the filter
should have a reasonably regular structure~ Also,
considering cost factors, it should have a minimum of
complexity. It is also a requisite that the number of
multiplications be kept to a minimum and it is preferable
that the wordlength of multiplier coefficients also be
kept to a minimum. There are other performance
considerations that lead to a need for increased signal
wordlength. Recursive filter structures tend to amplify
any quantisation noise which must inveriably occur during
the truncation operation that follows any multiplier
stage. It is usual to compensate for this amplification
by adopting additional bits in the signal wordlength in
order to keep filter noise at an acceptable level. The
lS signal amplitudes at internal nodes of the filter must
also be considered. A filter with a high effective
Q-factor may have signals at internal nodes as much as
40dB higher in amplitude than the input signal, at certain
frequencies, even if the overall gain is unity. Clipping
is prevented again by increasing signal wordlength.
By way of example, consider an application requiring
the~rejection of signals at 50Hz and 60Hz by at least 25dB,
rejection of all components at DC, with less than 0.7dB
attenuation of signals or 200Hz at a system sampling rate
Z5 of 8Hz.
B
4~3~3
--3--
A filter comprised of biquadratic sections would b~
very regular and easy to implement. However, such a
filter optimised for best noise and signal growth
performance, would include as many as fourteen 9-bit
multipliers and six delay elements. This would have a
noise amplification of approx. 13dB with noise gain at
internal nodes of OdB. The signal wordlength would n~ed
to be increased by an additional three bits fQr filter
compensation as aforesaid. It should howev~r be noted
that many alternative structures exist. Thus for example,
a filter with considerably worse signal growth
performance could be implemented using eleven multipliers
and five delay elements.
Disclosure of the Invention
It is intended to provide a digital filter of
reasonably regular structure, but one with less complexity
for equivalent or better performance than that provided
hitherto.
In accordance with the invention there is provided a
low frequency digital notch filter comprising:-
an input node;
an all-pass network first filter stage having an input
and an output, said input being connected to said input
node, the first filter stage including at least one first
delay element and at least one first coefficient
multiplier for multiplication by a coefficient K1, the
~3
~ ~84~3~
first delay element and first coefficient multiplier being
interconnected in such a way as to provide a transfer
function A(Z) for the first stage as follows:
A(Z)=(Z~n+Kl)/(l~K1Z n), where n~1;
a second filter stage having an input and an output,
the input of the second filter stage being coupled to the
output of the first filter stage, the second filter stage
including a second delay element and three second
coefficient multipliers for multiplication by coefficients
K2, K3 and K4 respectively, the second delay element and
second multipliers being so interconnected that the
transfer function B~Z) for the second stage is as follows~
B(Z)=t(K3+K2-K4)Z n 1]/(1-K3Z n), wherein n~l;
a filter output node coupled to the output of the
second filter stage; and
a feedforward line coupled between said input node and
said output node for summing the filter input with the
output of the second stage~ whereby to provide a notch
characteristic at a desired frequency.
The notch filter aforesaid may be combined in series
with a DC-rejection filter, according to the application.
According to a further aspect of the invention, there
is provided a T-section filter for use in a distal notch
filter as recited above, and including a delay element and
three coefficient multipliers for
~884~
- 4a -
multiplication by coefficients K2, K3 and K4 respectively,
the delay element and multipliers being so interconnected
that the transfer function B(Z) for said T-section filter
is as follows:
B(Z~=t(K3~K2-K4)Z n-l]/(l-K3z n~, where n~1;
and said T-section filter having a tapping node
located at the output of the delay element, such that the
transfer function C(Z) between the input to sai~ T-s~ction
filter and the tapping point is given as follows:
C(Z~ = K~Z n/(l-X3Z n), where n~1:
and a feedback line coupled between said tapping point
and said input.
Brief Introduction to the Drawinqs
In the drawings accompanying this application:-
Figure 1 is a circuit diagram of a low frequencydigital notch filter implemented in accordance with the
invention:
.; .,,
2 ~8~.3
5_
Figure 2 is a circuit diagram of a DC-rejection filter
for use in combination with the notch filter shown in the
preceding figure; and,
Figure 3 is a graph showing a typical gain frequency
response specification, and a frequency response attainable
using a combination of the filters shown in the preceding
figures.
Description of a Preferred Embodiment
An embodiment of the inven~ion will now be described,
by way of example only, with particular reference to the
drawings aforesaid.
The low frequency notch filter section of a composite
rejection filter is shown in figure l. Essentially, this
section is comprised of two sub-filters, an all-pass-
network filter l and a T-section filter 3.
The all-pass-network filter l incorporates a delay
element 5 and a coPfficient multiplier 7. The transform
function of this filter l has the form A(Z) given by the
expression:
AIZj = (Z-l + Kl) / (l + K1Z-l), where K is
the value of coefficient applied to the multiplier 7. It
may be implemented by a standard recursive structure - See,
for example, structures described in the article "Digital
Signal Processin~ Schemes for Efficient Interpolation and
Decimation" by RIA.Valerzuela and R.G.Constantinides,
3.,~
6 --
IEE Proc. Vo. 130 Pt.G No. 6 (Dec 1983) P. 232. The
all-pass-network filter ~, shown, includes branch nodes 9
and 11 between a common input, the delay element 5 and the
multiplier 7. It also includes an output node 13 which is
connected to the output of the delay element 5, and the
output of the multiplier 7. Cross-connections 15 and 17
respectively, are provid~d between the output o~ the delay
element 5 and the input of the multiplier 7 via branch
node 11, and between the output of the multiplier 7 and
the input of the delay element 5, via the branch node 9.
The T-section filter 3 is connected to the output node
13 of the all-pass-network filter 1. It is comprised of
three multipliers 19, 21 and 23 which provide multipli-
cation by coefficients K2, K3 and K4 respectively. The
transform functions B(Z) and C(Z) of this filter 3 have
the form given in the following expressions:-
B(Z)=[(Kl K4~K3)z 1-1]/[1-K3z 1]; and,
(input-to-output.)
--1 --1
C(Z)=K2Z /[l-K3Z ].
(input-to-tap.)
In the implementation of this filter 3, as shown in
figure l, the~multiplier 19, a branch node 27, the delay
unit 25 and the multiplier 21 are connected in series and
the delay unit 25 is shunted by the multiplier 23, via
~ ~B~3
the branch node 27. The output of the multiplier 21 is
connected to an output node 29. This latter node 29 is
also connected to the T-filter input for subtracting input
signal.
The low frequency notch filter section is completed by
an input node 31 and a terminal output node 33. A feedback
signal is extracted from the T-section filter 3 from a
point in the signal path between the delay unit 25 and the
multiplier 21, and subtracted from the input signal at the
input node 31. The output signal from the node 31 is fed
in parallel to the input of the all-pass-network filter 3
and to the output node 33, where it summed with the output
signal of the T section filter 3.
The low frequency notch filter section shown in figure
1 also includes a fifth multiplier 35 which is connected to
the output of the output node 33. This latter introduces
further multiplication by a coefficient K5, of value 1.
The DC-rejection filter section of the composite
filter is shown in figure 2~. This latter section is
connected in series with the low frequency notch filter
sectlon described above. Essentially this section is
comprised of an all-pass-network filter 1 and an output
node 37. The input signal to this section is fed in
parallel to the network filter 1 and to the output node
37. At this node 37 the input signal is subtracted from
1~84~33
--8--
signal from the output of the network filter 1. The
all-pass-network filter 1 may be of identical configuration
to that previously described with reference to figure 1.
The multiplier 7 in this case provides multiplication by a
coefficient K6. In the DC-rejection filter shown in
figure 2, the output node 37 is connected to a seventh
multiplier 3g. This multiplier 39 introduces further
multiplication by a coefficient K7 of value 1/2. This
choice of DC-rejection filter offers improvement
performance compared with its biquadratic section
counterpart, and is very similar in form to the notch
filter, thus improving the potential for efficient
implementation.
~ A typical performance specification is illustrated in
figure 3. Rejection at 50Hz is required and a level
response for frequencies above 200Hz. Suitable values for
the coefficients Kl to K7 are given in Table 1
appearing below:-
TABLE 1
20 Coefficient Value
Kl ~ [1-3/16]
K2 1/16
K3 1-1/128
K4 ~ 1/4
K5 1/2
K6 ~ [1-1/32]
K7 1/2
~-~8~34~33
As can be seen from figure 3, the required performance
is readily achieved when the coefficients given above are
adopted. The figure shows DC-rejection, lo~ frequencY
rejection at 50Hz, and, a near level response at
frequencies above 200Hz.
The following table gives the performance of the two
sections, figs 1 and 2, taken together.
TABLE 2
Structure: Reasonably regular, uses simple all-pass-network
as the basic building block for the filter.
Complexity: 7 multipliers, 3 delay elements.
Multipliers: 7 bit, although 4 of the 7 multipliers, being
simply a division by a power of two can be
very simply implemented.
Noise amplification: approx. lOdB.
Maximum noise gain at internal nodes: 3dB.
Signal wordlength: 3 additional bits required to compensate
for filter.
It can be seen that this stucture offers a number of
improvements over more conventional structures, at the
expense of a slightly less regular structure. It manages
to achieve similar performance to the biquadratic section
filter examined earlier for a giYen signal wordlength using
half the number of storage elements and half the number of
multipliers, and with the multiplications in many cases
being trivial, eg 1/2 etc.
3 ~884~33
--10--
The coefficients K1 to K7 may be varied to meet
the requirements of other applications, in particular to
vary the frequency of t'ne notch filter relative to the
samplin~ frequency~ One of the advantages of the stru~ture
aforesaid is that many of the multiplications may be
performed trivially ie. using divide-by-two cascades, even
when the coefficients are altered to conform to a different
requirement. Thus referring to figures 1 and 2, the
coefficients K5 and K7 would always be chosen equal to
1/2. It can usually ~e arranged for one if not both of the
coefficients K2 and K4 to be a power of 1/2 and the
coefficient K3 to be related to the coefficients K2 and
K4 as follows~ K3 = 1 - (2 * K2 * K4).