The present invention relates in general to switched bandwidth
digital filters using swapped coefficient sets, and more specifically, to slewing
a subset of the coefficients to reduce audible effects during switching.
Digital filters manipulate discrete samples of an input
signal to produce a filtered output signal. Various filter structures are known
in the art, such as those for finite impulse-response (FIR) filters and infinite
impulse-response (IIR) filters. Higher order IIR filters (providing more selectivity)
are typically implemented using a plurality of lower order filters connected in
During processing of a signal, it may become necessary
to change the filtering of the signal (e.g., a change in bandwidth, passband characteristic,
group delay, or other filter parameters). To minimise hardware and/or software requirements,
it is desirable to use the same filter structure before and after the change by
merely changing the digital filter coefficients.
In a digital signal processing (DSP) radio receiver, for
example, a digital channel filter is applied to an intermediate frequency (IF) signal
to select the desired signal and reject other broadcast signals. A wide or a narrow
passband may be used for the channel filter depending upon the presence of interfering
adjacent or alternate channel broadcast signals. When switching between the two
bandwidths by switching between two coefficient sets in a DSP filter, the sudden
change in bandwidth may be noticeable to the listener and may lead to a perception
of poor quality.
It would be possible to gradually slew the bandwidth between
wide and narrow, but the computation and/or storage requirements for updating all
the coefficients of the filter during slewing would be too great and would be expensive.
WO 97/39526A discloses the switching between a first and
second digital FIR filter in which sections of each filter are rendered inoperative
during the switching.
According to the invention, there is provided a method
for slewing between a first bandwidth characteristic and a second bandwidth characteristic
in a switched bandwidth digital filter including a plurality of second-order filter
sections. Each of the sections comprises respective coefficients having a first
set of values which provides the first bandwidth characteristic and having a second
set of values which provides the second bandwidth characteristic. The coefficients
of a selected filter section are slewed in a plurality of iterative steps between
the first set of values and a surrogate set of values while coefficients in the
filter sections other than the selected filter section are held constant. The coefficients
of the filter sections other than the selected filter section are swapped from their
respective first set of values to their respective second set of values. When going
from wideband to narrowband, the slewing step may preferably be performed before
the swapping step. For going from narrowband to wideband, the swapping step may
preferably be performed before the slewing step. In each case, the slewing/swapping
order might sometimes be reversed, depending on analysis and/or subjective listening,
and depending on how easy it is to find a good surrogate set of values.
The selected filter section is preferably the one section
which has the most critically damped characteristic (i.e., most flat within the
passband). The surrogate coefficients of the selected section are designed, in combination
with the remaining unchanged sections, to provide a passband characteristic which
approximates an overall passband of said digital filter when it is providing its
narrower bandwidth characteristic.
The present invention has the advantage of eliminating
a sudden bandwidth change without slewing a large number of coefficients within
The invention will now be described further, by way of
example, with reference to the accompanying drawings, in which:
- Figure 1 plots reception field strength in a local reception area in which adjacent
channel interference exists for a desired radio channel of interest;
- Figure 2 is a block diagram showing portions of a DSP radio receiver;
- Figure 3 is a block diagram showing DSP processing of an intermediate frequency
signal as used in the present invention;
- Figure 4 shows the internal structure of a channel filter;
- Figure 5 is a flowchart showing the switching of the filter bandwidth from wide
- Figure 6 is a flowchart showing the switching of the filter bandwidth from narrow
to wide; and
- Figures 7A-7D show sequential filter section spectra according to a coefficient
set for providing a wide passband, which forms a basis for selecting a filter section
Figure 1 shows a frequency spectrum 10 of a desired radio
broadcast having a centre frequency 11 and occupying an assigned channel fd
between a lower frequency 12 and an upper frequency 13. An upper adjacent channel
fu is shown containing a broadcast signal 14 with substantially no excess
signal content in the desired frequency channel, whereby the upper adjacent channel
is not causing interference. However, a lower adjacent channel at f1
is shown to include a radio broadcast having a frequency spectrum 15 including significant
signal content above lower frequency 12 of the desired channel. The resulting interference
degrades reception of the desired radio broadcast.
Adjacent channel interference can be reduced by means of
narrowing the passband of a bandpass filter in the receiver to reduce the signal
content from the adjacent channel passing through the receiver. Thus, Figure 1 shows
a narrow bandwidth 16 which can be switched into the intermediate frequency signal
path to alleviate adjacent channel interference. When no adjacent channel interference
is present, a wide bandwidth 17 is used in order to maximise quality of the received
desired signal. Within the receiver, centre frequency 11 is translated to an intermediate
frequency, which may be a zero intermediate frequency. In that case, the filter
is a lowpass filter.
Figure 2 is a block diagram showing a radio receiver using
digital signal processing. An antenna 20 receives broadcast RF signals which are
coupled to an RF amplifier 21. Amplified RF signals are provided to one input of
a mixer 22. A local oscillator 23 provides a mixing signal to a second input of
mixer 22, the mixing signal having a frequency under control of a tuning control
circuit (now shown). A carrier-based signal in the form of an intermediate frequency
(IF) signal is provided from mixer 22 to the input of an analogue-to-digital (A/D)
converter 24. A digitised IF signal is provided to digital signal processor (DSP)
block 25 for filtering, demodulating and other further processing of the resulting
audio signal. Block 25 includes data memory and program memory for performing these
functions. A final audio signal is output from DSP-25 to the input of a digital-to-analogue
(D/A) converter 26 which provides analogue audio signals to a speaker system 27.
Processing of the digitised IF signal within DSP 25 is
shown in greater detail in Figure 3. The embodiment of Figure 3 is particularly
adapted for processing AM signals. The digitised IF signal is provided to the input
of a complex mixer 30 to produce in-phase (I) and quadrature-phase (Q) signals.
An oscillator 31 produces an injection signal fif which is nominally
equal to the intermediate frequency of the IF signal so that the IF signal is mixed
to a new IF frequency of approximately zero Hertz. The injection signal is coupled
directly to one input of a first mixer 32 and through a 90° phase-shift block
33 to an input of a second mixer 34. The digitised IF signal is provided to respective
inputs of mixers 32 and 34 to generate the I and Q signals. The I and Q signals
are decimated by decimate blocks 35 and 36, respectively, to provide sample-rate-reduced
signals to the inputs of channel filters 37 and 38.
Other non-zero IF frequencies or non-complex signal representations
can be used with the present invention. However, a zero-IF complex representation
has many advantages in DSP processing such as compact code size, minimised chip
area, and efficient data manipulation.
Channel filters 37 and 38 can be loaded with a coefficient
set #1 or a coefficient set #2 through a multiplexer 40 under control of an adjacent
channel detect block 41. One coefficient set provides a wide bandwidth while the
other coefficient set provides a narrow bandwidth. A coefficient set to be used
at any one time is selected depending upon the presence of adjacent channel interferers.
Using a zero Hertz intermediate frequency, channel filters 37 and 38 are implemented
as lowpass filters with the narrower filter having an upper cut-off frequency which
is lower than that of the wider filter. The presence of adjacent channel interferers
can be detected using any conventional method as known in the art. The filtered
outputs of channel filters 37 and 38 are provided to a signal detector 42 for generating
an audio output which may include left and right stereo signals, for example.
The memory blocks for coefficient sets #1 and #2 may further
contain coefficient values to be used during slewing of a selected one of the filter
sections. Alternatively, information may be stored to allow for these slewing coefficient
values to be calculated when needed.
The channel filters may be implemented using various filter
structures and types. An infinite impulse response (IIR) filter as shown in Figure
4 will be described herein as an example since its use is desirable for its advantages
of compact size.
Figure 4 shows a typical architecture for an IIR filter
comprising three second-order sections cascaded in series between an input x(n)
and an output y(n). A filter may include an overall gain term G0 applied
to one input of a multiplier 45 which receives input x(n) at its other input. Alternatively,
the gain term Go may be distributed to the individual sections as is known in the
Second order sections 46, 47 and 48 are connected in series
to produce a sixth order filter. First section 46 includes a multiplier 50 for multiplying
the input to section 46 by a coefficient b
0,1. The resulting product is provided to one input of a summer 51. The
output of summer 51 provides the output node for section 46, and is also an internal
node of the whole filter.
The input to section 46 is delayed by one sample period
in a unit delay block 52 and then input to a multiplier 53. Coefficient
1,1 is applied to a second input of multiplier 53 and the output is provided
to summer 51. The output of unit delay block 52 is passed through a further unit
delay in unit delay block 54 prior to multiplying it in a multiplier 55 by a coefficient
2,1. The output of multiplier 55 is provided to yet another input of
summer 51. The b coefficients provide the feedforward terms for section 46.
Section 46 also includes feedback terms wherein the output of summer 51 is delayed
in a unit delay block 56. The delayed output is coupled to a multiplier 57 which
also receives a coefficient a
1,1. The output of multiplier 57 is coupled to another input of summer
51. The delayed output from unit delay block 56 is passed through a unit delay block
58 and then to an input of a multiplier 59. Coefficient a
2,1 is supplied to another input of multiplier 59 and the resulting product
is coupled to summer 51. The output of summer 51, comprising the internal node for
the first section 46, is coupled to the input of second section 47. Sections 46
and 47 are shown overlapping since unit delay blocks 56 and 57 are shared between
the two section operations in order to minimise hardware requirements.
The input of section 47 (from summer 51) is applied to
a multiplier 60 which also receives coefficient b
0,2. Additional b coefficients b
1,2 and b
2,2 are applied to multipliers 62 and 63, respectively, and the resulting
products are added in a summer 61. Unit delay block 64, multiplier 65, unit delay
block 66, and multiplier 67 provide feedback terms using coefficients
1,2 and a
2,2, as in the previous section. Section 48 operates in the same manner
using b coefficients for the third section b
1,3 and b
2,3 and a coefficients a
1,3 and a
2,3. In order to provide a final filtering of higher order, more second
order sections may be cascaded in series after section 48.
Various methods are well known for determining appropriate
values for the a and b coefficients. In a preferred embodiment, a
Butterworth filter structure is used. First and second coefficient sets are separately
derived for providing a first bandwidth characteristic and a second bandwidth characteristic,
respectively, in order to switch a filter between bandwidths while using the same
filter hardware and structure as shown in Figure 3. Thus, by swapping coefficient
values between the two coefficient sets, either the wide bandwidth or the narrow
bandwidth channel filter can be selected. As used herein, a coefficient set refers
to all the a and b coefficients for all the filter sections for providing one bandwidth
Using the coefficient sets as obtained with the conventional
process, significant audible effects may occur in the audio output when the filter
coefficients are changed. The present invention gradually changes the overall filter
characteristic to eliminate the audible change in frequency response.
If the bandwidth were to be slewed in discrete steps by
using conventional coefficient design techniques at a number of intermediate bandwidths
between the wide and narrow settings, a large amount of memory and/or computational
resources would be required. If 50 intermediate steps were used, then 50 different
full coefficient sets would have to be stored or calculated, making the slewing
function an expensive one to implement. However, applicant has found that by slewing
one specially selected filter section in a particular way, the desired result can
still be achieved.
Slewing all coefficients in all filter sections would cause
both the passband and stopband of the filter to gradually change. In the radio receiver
of the preferred embodiment, this is not needed. A bandwidth change from narrow
to wide occurs only if the receiver has determined that no spurious signals exist
that will be in the passband after slewing is completed. In that case, what happens
in the stop band does not have a significant audible impact, but the passband does.
A bandwidth change from wide to narrow occurs when spurious signals are present
in the wider bandwidth. There will be a more severe, audible effect when switching
the stopband, but this is less important since the audible reception condition in
this case will not be good anyway, and the slewing of the passband will still be
To slew the corner frequency of a single second order section
of an IIR filter, the a
1 term is slewed. In order to preserve a constant overall passband gain
of the section, the b
1, and b
2 terms must be adjusted accordingly. In a Butterworth filter,
2, so only b
0 must be calculated. Slewing a
1 is accomplished by adjusting it slightly (by an adjustment amount)
on each iteration of a timing loop, until a
1 is completely changed from its start value to its end value. On each
iteration, the b values are determined based on the current value of
1. Alternatively, some or all the slewing values could be stored in a
table. Note that by changing a
1 without changing a
2, the corner frequency of the section is changed without changing the
Q of the section.
The start values (when going from narrow to wide) or the
end values (when going from wide to narrow) for the slewing of the selected section
may correspond to the actual set of coefficients for the selected section when switched
to the narrowband characteristic. However, it may be more desirable to use a surrogate
set of coefficient values which provides a passband characteristic (when combined
with the remaining sections) which best approximates the overall passband of the
digital filter when it is providing its narrowband characteristic. By using the
best approximation within one section of the overall filter passband characteristic,
any audible effects of the sudden switching of the filter coefficients all together
are reduced as much as possible. The surrogate values are determined during filter
design using conventional techniques.
Referring to Figure 5, the procedure for slewing from wideband
to narrowband begins in step 80 where coefficient a
1 of the selected section is changed by an adjustment amount. The total
time to slew equals the number of iterations divided by the iteration rate, and
the total span of values equals the number of iterations times the adjustment amount.
A typical slewing time may be about 12 seconds and a typical number of iterations
may be about 123. These are not necessarily the same in both slewing directions.
In step 81, the b coefficients of the selected section
are calculated and loaded into the filter along with the new value for
1. Specifically, the new value for b
0 is determined as k1*a1+k2, where k1
and k2 are predetermined constants. New values for b
1 and b
2 are determined from b
0 as given above. The a
2 term need not be changed or slewed since it has no effect on the corner
frequency of the filter section.
In step 82, a check is made to determine whether the surrogate
value of a
1 has been reached. If not, then a loop delay is implemented in step
83 and then a return is made to step 80 for the next iteration. If the surrogate
value of a1 has been reached, then a coefficient swap to the narrow filter
coefficient values is made in step 84. If the surrogate values are not equal to
the narrow values for the selected section, then the selected section also participates
in the swap.
Referring to Figure 6, the procedure for slewing from narrowband
to wideband begins in step 90 where all coefficients (except the selected section)
are swapped to their wide values. If the surrogate values do not equal the narrow
values for the selected section, then the selected section is swapped to the surrogate
values in step 91.
In step 92, coefficient a1 of the selected section
is changed by the adjustment amount. In step 93, the b coefficients of the
selected section are calculated and loaded into the selected section along with
the new value for a
1. In step 94, a check is made to determine whether the end wideband
values have been reached. If not, then a loop delay is implemented in step 95 and
then a return is made to step 92 for the next iteration. If the wideband values
have been reached, then the procedure is done.
A preferred criteria for selecting the filter section to
be slewed will be described with reference to Figures 7A-7D. For most applications,
it is desirable to choose the filter section with the flattest frequency response
in the passband. A section with an optimally flat frequency response is one that
has a Q value of 0.707, or is critically damped. Thus, it is a preferred embodiment
of the invention to select the filter section having a Q value in its wideband configuration
closest to 0.707 as the section to be slewed. The selected section may well also
be the one with a Q value closest to 0.707 in its narrowband configuration, but
this is not required since surrogate coefficient values can be used for slewing.
Figures 7A to 7D show frequency responses for individual sections of the filter.
Figure 7B corresponds to the flattest frequency response and has a Q value closest
to 0.707. A section with a lower Q value (e.g., Figure 7D) has a shallower roll-off
and is not flat over its passband (considered as a separate filter). If such a section
is the one slewed, the whole filter's passband becomes less flat and the corner
frequency is not well controlled during slewing. A section with a higher Q value
is likewise not flat in its passband (e.g., Figure 7A) and is less able to control
the corner frequency during slewing.
A filter section other than the flattest may alternatively
be used for slewing. For example, there may be applications where it is desirable
to either emphasise or de-emphasise particular frequencies during slewing, for which
an appropriate filter section can be selected.