Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-1_
IMPROVEMENTS RELATING TO FREQUENCY SYNTHESISERS
FIELD OF THE INVENTION
This invention relates to sigma-delta modulators and particularly but not
solely to
radio frequency synthesisers which use these modulators in fractional
division.
More particularly the invention relates to modulator systems which are able to
provide an extended range of fractional division ratios and reduced emission
of
spurious frequencies.
BACKGROUND TO THE INVENTION
Radio communication devices employ frequency synthesisers in various ways to
control transmission and reception of signals. A synthesiser generally
includes a
reference oscillator which generates a stable reference frequency signal and
is used
to determine the output of a frequency controlled oscillator which in turn
generates
a variable RF output signal. This output signal is generally coupled to an
antenna
of the communication device by way of one or more mixers which modulate or
demodulate the signal for transmission or reception respectively. The
synthesiser
is programmed by a control unit such as a digital processor to produce the
controlled
oscillator signal at a range of frequencies as required by the device.
Frequency
synthesisers are also often used in power amplification systems for radio
transmitters.
Indirect frequency synthesisers use one or more phase lock loops (PLLs) to
generate
the variable output signal from the frequency controlled oscillator. The phase
lock
loop contains a phase discriminator which generates an output according to the
phase difference between the reference signal and a feedback signal. The
feedback
signal is generally produced by dividing the frequency of the output from the
controlled oscillator. Output from the phase discriminator is applied to a
loop filter
which provides a control signal for the controlled oscillator. Voltage rather
than
current controlled oscillators are normally used. In general terms, a feedback
loop
of this kind attempts to match the frequency of the controlled oscillator to a
multiple
of the reference frequency and stabilise with a predetermined phase difference
between the reference and feedback signals.
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-2-
Frequency division of the output from the frequency controlled oscillator can
be
implemented in various ways to enable a relatively low frequency reference to
determine a wide range of variable RF output frequencies. Integer division
techniques are acceptable in many circumstances. However, fractional
techniques
are becoming common and allow the synthesiser to achieve almost arbitrarily
fine
frequency resolution. These techniques modulate the instantaneous integer
divide
ratio of the feedback to the phase discriminator to produce average non-
integer
division ratios. The non-integer values have an integer part IP with value N,
and a
fractional part FP with a value typically in the range {0,1 }. Cyclic
variation of the
division value generally produces spurious frequencies and additional phase
noise
in the synthesised output signal. Various cancellation schemes such as phase
interpolation have been employed to reduce the fractional spurs and noise but
generally require an increase in complexity and cost of the synthesiser to
achieve
significant reduction in the amplitude of the spurs.
Fractional-N synthesisers which use sigma-delta modulation to reduce phase
noise
and spurs resulting from non-integer division values are well known, as
described
in US 4,609,881 for example. This modulation technique arose as a development
in analog-to-digital conversion and has since been widely used in electronic
communication devices for a range of purposes. It involves feedback to improve
the
effective resolution of a coarse quantiser and allows shaping of the noise
which
arises from quantisation. In general terms, input is fed to the quantiser via
an
integrator with the quantised output being fed back and subtracted from the
input.
The output of the modulator therefore contains the original signal plus the
first
difference of the quantisation error. A first order process of this kind may
be
implemented wholly or partly using digital signals. A detailed discussion of
sigma-
delta techniques can be found in Delta-Sigma Data Converters, IEEE Press 1997.
Higher order sigma-delta modulators generally use two or more integrators each
receiving feedback from the output to improve the overall noise performance. A
cascade is also sometimes used whereby the output of two or more modulators is
combined in a way which cancels the noise that they individually produce. In a
cascade of two first order modulators for example, output from the integrator
of the
first modulator is fed to the second modulator. The output of the second is
differentiated and subtracted from the output of the first to provide a net
signal. This
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-3-
leaves the noise as the second difference of the quantisation error of the
second
modulator, in a form similar to that of a second order modulator. Mufti-level
quantisers have also been used to improve the stability of higher order and
cascaded
modulators.
SL)TvIMARY OF THE INVENTION
It is an object of the present invention to provide for improved performance
of
fractional-N frequency synthesisers. The improvements are particularly
directed to
increased flexibility in the selection of fractional division values and
decreased
output of spurious frequencies or noise. In general these improvements are
enabled
by a sigma-delta modulator having a quantiser with logic control in the output
and/or
feedback stages.
Accordingly in one aspect, the invention may broadly be said to consist in a
frequency synthesiser comprising: a synthesiser loop including a programmable
frequency divider, and a controller which determines average fractional
division
values for the divider, wherein each division value comprises selected integer
and
fractional parts, and the selected fractional part varies over a range other
than {0,1 }.
In a second aspect the invention may also be said to consist in a frequency
synthesiser comprising: a synthesiser loop including a programmable divider, a
divider control means which provides a signal to the divider to cause
fractional
division and thereby produce a required output frequency from the loop,
wherein the
control means includes at least one digital modulator having an adder, a latch
coupled between an output and input of the adder, and an output logic stage
which
carries out a logical operation on one or more bits which are output by the
adder.
In a third aspect the invention may further be said to consist in a frequency
synthesiser comprising: a synthesiser loop including a programmable divider, a
divider control means which provides a signal to the divider to cause
fractional
division and thereby produce a required output frequency from the loop,
wherein the
control means includes at least one digital modulator having an adder, a latch
coupled between an output and input of the adder, and a feedback logic stage
which
carries out a logical operation on one or more bits which are output by the
adder.
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-4-
In a fourth aspect the invention may be said to consist in a digital modulator
comprising: an adder which receives a control signal and an error signal as
inputs
and produces a mufti-bit output by addition of the inputs, an output logic
stage which
receives an input derived from at least one bit of the output of the adder and
carries
out a logical operation on that input to produce a modulator output signal,
and a
feedback path including a latch which combines at least two groups of bits
from the
output of the adder to produce the error signal.
In a further aspect the invention may be said to consist in a modulator
arrangement
comprising: two or more modulators each having an adder, a latch coupled
between
an output and an input of the adder, and an output logic stage, wherein the
modulators are connected so that output of the latch in each modulator is
coupled to
an input of the adder in a subsequent modulator, if any, and the output of the
logic
stage in each modulator is coupled to a common combination stage to produce an
output for the arrangement.
BRIEF DESCRIPTION OF DRAWINGS
Preferred embodiments of the invention will be described with respect to the
accompanying drawings, of which:
Figure 1 schematically shows a frequency synthesiser having a frequency
divider which is programmable with integer division values,
Figure 2 shows a frequency synthesiser in which the divider is modulated for
fractional division,
Figures 3a, 3b respectively show a conventional accumulator performing as
a sigma-delta modulator and a corresponding quantiser transfer function,
Figures 4a, 4b respectively show the Z-transform model of an ideal sigma-
delta modulator and a transfer function corresponding to an ideal two-level
quantiser,
Figure 4c shows a Z-transform model equivalent to that of Figure 4a for
comparison with Figure 6,
Figure 4d is a further model indicating how dither signals may be added to
a sigma-delta modulator to reduce limit cycles,
Figure 5 shows a three stage sigma-delta modulator formed by a cascade of
first order modulators such as shown in Figure 3a,
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-5-
Figure 6 is an accumulator circuit with logic stages forming an improved first
order sigma-delta modulator according to the invention,
Figures 7a, 7b, 7c demonstrate two, four and eight-level transfer functions
which may be implemented by the quantiser in Figure 6,
Figure 8 is a table showing how the logic stages of the modulator in Figure
6 can be implemented to form a two-level quantiser,
Figures 9a, 9b are tables showing how the logic stages of the modulator in
Figure 6 may be implemented to form a four-level quantiser,
Figure 10 indicates fractional division values produced by a modulator having
a quantiser with a four-level transfer function such as shown in Figures 9a,
9b,
Figure 11 shows a three stage modulator formed by a cascade of first order
modulators according to Figure 6 each possibly having different quantiser
levels,
Figure 12 is a table indicating scale factors required for input and output to
the individual modulators in Figure 11 when the quantisers have different
output
levels,
Figure 13 is a table summarising properties of modulators for a range of
possible mufti-level quantisers,
Figures 14a, 14b respectively show sample output from the mufti-stage
modulator systems of Figures 5, 11,
Figure 15 is an alternative modulator based on Figure 6 in which feedback
from the quantiser is dithered, and
Figure 16 is a table indicating a possible implementation of the quantiser.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
Referring to these figures it will be appreciated that a frequency synthesiser
according to the invention or more particularly a sigma-delta modulator can be
constructed in various ways within the scope of the claims. The preferred
embodiments are described by way of example only. The known components of
synthesiser and modulator devices will be understood by a skilled person and a
detailed explanation of their function need not be given.
Figure 1 shows a simple frequency synthesiser using a phase lock loop (PLL) 10
as
a feedback control system. The loop contains a phase discriminator 1 l, a
filter 12,
reference oscillator 13 and a voltage controlled oscillator (VCO) 14. A
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-6-
programmable frequency divider 15 is included in a feedback path 16 from the
VCO
to the phase discriminator. The discriminator acts to compare the frequency fr
of the
signal from the reference oscillator with the frequency fo of the signal from
the VCO
after division by an integer value N in the divider. If the phase of the
reference
signal leads that of the divided VCO signal then the output of the
discriminator is
effectively a frequency down signal which serves to decrease the frequency of
the
VCO. If the phase of the reference signal lags that of the divided VCO signal
then
the output of the discriminator is a frequency up signal which increases the
frequency of the VCO. In normal operation the loop stabilises so that the
signals
input to the discriminator have zero phase difference and the output of the
VCO
forms the output of the synthesiser with fo = N fr . A controller 17 sets N
within a
range of discrete integer values.
Figure 2 shows a conventional fractional-N frequency synthesiser in which the
controller 17 varies the instantaneous value of N by way of a modulator 20.
This
creates a range of non-integer division values which are available from a dual
modulus divider 25 in the feedback loop 10. The division value is periodically
altered from N to N+1 during a cycle determined by a controller 27. In this
example
the modulator simply involves a K-bit accumulator which receives a control
word
k from the controller on line 21. Each output pulse from the divider clocks
the
accumulator on line 22 and adds the input word to the accumulator contents. If
the
contents exceed 2K then an overflow signal is generated on line 23 and causes
a
single division by N+1 rather than N in the divider. For a constant input word
the
accumulator will overflow on every 2"/k clock pulses. This creates an average
division value having an integer part IP = N and a fractional part FP = k/2K <
1
which are determined by the controller. In normal operation the loop
stabilises so
that the signals input to the discriminator has an average phase difference
equal to
zero and the output of the VCO forms the output of the synthesiser with fo =
(N +
k/2") fr .
Figure 3a shows the accumulator of Figure 2 in more detail. An adder 30 has
two
inputs one of which receives the control word k from the controller 27 on line
21.
The second input receives the output signal c of the adder through latch 31.
An
overflow signal from the adder is passed to the divider on line 23 while the
latch is
clocked by output from the divider on line 22. The accumulator acts as a first
order
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
_ '7 _
sigma-delta modulator in which the overflow signal represents quantisation Q
of the
output signal c of the adder. Figure 3b demonstrates a quantiser transfer
function
Q(c) which is not symmetrical about zero.
Figure 4a gives the Z-transform model of an ideal first order sigma-delta
modulator
by way of comparison with the modulator of Figure 3 a. The input k is fed to a
two-
level quantiser 40 by way of an integrator represented by sum function 41 and
delay
42. Output Y of the modulator is fed back through delay 43 and subtracted from
the
input by sum function 44. Output of the integrator is represented as a signal
a
approximately comparable with the signal c at the second input of the
accumulator
in Figure 3a. Figure 4b demonstrates the ideal two-level quantiser transfer
function
Q(u) which is symmetrical about zero. Figure 4c is a model equivalent to that
of
Figure 4a which is mentioned later in relation to Figure 6.
Figure 4d indicates how a dither signal may be added in a modulator
arrangement.
Dithering is a known process in which typically a random or at least pseudo-
random
sequence is added to a digital data stream to reduce the likelihood that any
cyclical
pattern will develop. The mean value of the sequence is zero so there is no
overall
effect on the output. In the present case a cyclical output from the modulator
will
be passed to the frequency divider and cause spurious frequencies in the
output of
the synthesiser. If required in a particular implementation a dither sequence
d may
be added either at the input to adder 44 through an extra sum function 48 or
at the
input to quantiser 40 by way of sum function 49. This randomises the
quantisation
noise signal a more effectively. The sequence d must normally be pre-filtered
to
preserve noise shaping by the modulator. preferably by a transformation (1-
z').
Figure 5 shows a three stage modulator 50 formed by a conventional cascade of
first
order modulators 51, 52, 53 such as the accumulator shown in Figure 3a. A
control
word X produces a relatively complex signal Y which may be provided for the
divider 25 in Figure 2 to determine the average division value of the feedback
loop
10. The contents of each modulator stage forms an error signal which is
provided
as an input to the next stage while each output is passed through a pair of
respective
delay elements 54. A selection of outputs from the modulators and delay
elements
is passed to a combination stage ~~. These outputs are selected and combined
so
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
_$_
that the signal Y contains a higher order correction for quantisation errors.
Each
modulator and each delay circuit is clocked by the output of the divider.
Fractional-N division by way of an accumulator overflow arrangement as shown
in
Figure 2 leads to spurious frequencies in the output of the VCO 14. These
spurs are
offset from the output frequency fo by nfrk/2" where n = 0, l, 2, 3, ... . For
large
values of n the spurs fall outside the bandwidth of the loop 10 and are
attenuated by
the filter 12. Additional circuitry is normally required to reduce those spurs
which
lie within the bandwidth and this increases the complexity and cost of the
synthesiser. Division modulated by output from the cascade in Figure 5 is free
of
fractional spurs for most control words X. However, the signal Y now takes a
range
of integer values and the divider 25 must be capable of performing multiple
modulus
division in the loop. In both cases the fractional part FP of the average
division
value is limited to lie in the range f 0,1 } for any given integer part IP.
Figure 6 shows a preferred modulator 60 according to the invention which may
be
used to extend the range of the fractional part FP of the average division
value in a
frequency synthesiser. A control word X produces an output Y which may be used
to modulate the divider 25 in Figure 2. All signals in this figure are digital
and are
in two=s compliment format. An n-bit adder 61 has two inputs one of which
receives the control word. The second input receives an error signal a derived
from
output of the adder after various feedback processes applied to groups of most
and
least significant bits. An output logic stage 62 receives a group t of msbs
from the
adder 61 and operates on the bits during a quantisation process which produces
the
modulator output. A feedback logic stage 63 also receives the group t from
adder
61 and operates on the bits in a feedback process which determines overload
and
stability performance of the modulator. An m-bit adder 64 receives a group of
m
msbs from the adder 61 and a group of m bits from the feedback logic stage 63.
A
latch 65 receives a group of n-m lsbs from the adder 61 and a group of m bits
from
the adder 64 to form the error signal. Latch 65 receives a clock signal which
moves
the modulator from one state to the next through addition processes in each of
the
adders 61 and 64.
The output and feedback logic stages 62 and 63 in Figure 6 may be provided in
various ways. A dedicated Boolean operation on the bit group t is one
approach.
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-9-
Alternatively either stage might be a multiplexer from which the value of bit
group
t selects the output. The output values of the multiplexers may be set in
hardware
or held in registers for example.
A dither signal is readily added to the modulator in Figure 6 if required,
either at the
input to adder 61 with control word X, at the output from the adder 61, or as
part of
the feedback logic stage 63 as described below. A dither signal will generally
increase the amplitude of the signal to which it is added. Given the increased
fractional range of the present modulator an extra adder equivalent to sum
function
48 in Figure 4d would be necessary to add a dither signal before the input of
adder
61. An extra adder equivalent to function 49 would be required for addition at
the
output of adder 61.
Figures 7a, 7b, 7c demonstrate ideal transfer functions Q(u) for two, four and
eight
level quantisers respectively, each of which may be implemented by the
modulator
60 of Figure 6. Each function produces a multiple level output signal f
according
to where the input signal a lies within one of a series of decision regions
defined by
points d. In general the range of possible decision regions and corresponding
levels
in the output signal is determined by the number of bits in group t from the
adder 61.
One, two and three bit groups provide two, four and eight levels of
quantisation
respectively, for example. That these implementations are possible is
indicated by
a comparison of modulator 60 with the model of an ideal sigma-delta modulator
in
Figure 4c. Adder 61 performs the sum function 44. Output stage 62 forms the
quantiser 40. A digital-to-analog function implicit in the feedback of Figure
4c is
performed by feedback logic stage 63 in order to allow this comparison. Adder
64
performs the sum function 46. Latch 65 performs the delay function 47. The
properties of the quantiser 40 are determined by an appropriate selection of
parameters within the output logic stage 62.
Figure 8 is a table outlining a selection of parameters for the output and
feedback
logic stages which will implement an ideal rivo-level quantiser in the
modulator of
Figure 6. In this table B=2° where n is the length of adder 61 and
group t contains
a single bit. Column (i) contains possible values of t which are required to
produce
the two levels. Columns (ii) and (iii) give the corresponding decision regions
in
alternative decimal and binary forms. Columns (iv) and (v) set out
functionality of
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
- 10-
the output logic stage 62 which in this simple case acts as an inverter. A
more
complex mapping of a multiple bit group t to quantiser output bits is required
for
higher levels. Columns (vi), (vii), (viii) set out functionality of the
feedback logic
stage 63. Column (vi) contains bits which determine the feedback ratio and can
be
chosen with some freedom. Columns (vii) and (viii) give the ratios for outputs
according to column (v) in binary and decimal forms.
Figures 9a, 9b are tables outlining alternative selections of parameters for
the output
and feedback logic stages which will implement an ideal four-level quantiser
in the
modulator of Figure 6. The fractional part FP of the division value in this
example
ranges over {-1.5, 1.5} as shown in Figure 10. Figure 9a is an example
indicating
a selection for decision regions which give optimum noise performance. Figure
9b
is an example indicating a selection having fewer bits for optimum hardware
e~ciency. In each table the left hand portion represents decision regions for
input
to the quantiser, the middle portion represents n-bit output by the adder 61,
and the
right hand portion shows feedback ratios f and output signals Y. In general
for any
implementation the choice of feedback ratios affects overload performance and
therefore noise shaping properties of the modulator for particular values of
the
control word X. A high overload point allows a wide range of control values
but
degrades the performance requiring a balance of considerations.
Figure 10 demonstrates a range of fractional division values which may be
produced
in a synthesiser by a modulator such as outlined in Figures 9a, 9b when
implementing a four level quantiser. In this example the value of FP ranges
over {-
1.5, 1.5} about an integer part IP which is shown with three successive
values: N-1,
N, N+1. The corresponding values of FP are: 1.25, 0.25 and -1.25, so that the
sum
of IP and FP equals N + 0.25 in each case. It can be seen that this average
division
value can be achieved in three different ways by an appropriate selection of
FP for
any one of the three different values of IP. Still greater ranges of FP are
possible
using other parameters in the output logic stage 62 of the modulator. Values
of IP
and FP may be selected with greater flexibility than possible in conventional
synthesisers where the value of FP is generally limited to {0,1 } at best. The
extended range for FP provides an advantage in situations where a change in
frequency of the synthesiser would conventionally require a change in both the
integer and fractional parts of the average division ratio. Changes in
frequency
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-11-
within the broader range possible for the value of FP can now be accommodated
by
a change in the fractional part alone. Less information regarding frequency
changes
need be programmed in the controller 17 of a frequency synthesiser
incorporating
the preferred modulator.
Figure 11 shows a three stage modulator 110 formed by a cascade of first order
modulators 111, 112, 113 such as that shown in Figure 6. Overall the control
word
k produces a complex signal Y which may be provided for the divider 25 in
Figure
2 to determine the fractional part FP of the average division value in the
feedback
loop 10. Each modulator may have a different quantisation function, and in
this
example they are seen to have 1, 2, 3 bit outputs respectively. This
arrangement is
generally similar to that of Figure 5 except that scaling functions must now
be
included so that the output levels of the respective quantisers are combined
correctly.
The example is simplified in that the modulators are each assumed to have an n-
bit
adder 61. Output of modulators 111, 112, 113 are scaled by S 12, 522, S32
respectively before passage through the delay elements 114 to the combination
stage
115. The inputs of modulators 112, 113 may also be scaled by 521, S31. In
general
the scaling is dependent on the capacity of the adder 61, the first
quantisation step
and the overload point of each modulator.
Figure 12 is a table outlining a selection of scale factors for possible
implementations of the modulators in Figure 11, assuming that the adders 61
shown
in Figure 6 all have equal length. Overall resolution of the arrangement is
determined by the first stage which must have sufficiently high capacity.
Capacity
of the later stages can be reduced without substantial loss of noise shaping
ability.
A preferred arrangement is shown in the fourth row of the table, where a
modulator
having a quantiser with 4-levels is followed by two each with 2-levels.
Scaling by
a power of 2 is straightforward in binary signals and by careful design can be
done
without additional hardware.
Figure 13 is a table outlining some properties of a small number of possible
multi-
level modulators of the kind shown in Figure 6. The table includes an
indication of
the range of possible output levels in a third order modulator such as shown
in
Figure 11 when formed by an arrangement of first order modulators having
identical
properties. Ranges of up to f -3.5, 3.5 } in the fractional part of the
division ratio
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
- 12-
which can be achieved in a frequency synthesiser are indicated, by way of
example.
Still higher ranges are possible with other implementations. This table
includes
modulators having both even and odd numbers of quantisation levels and with a
linear relationship between decision regions and feedback levels. Non-linear
relationships can also be created if required to avoid output of particular
spurs.
Figures 14a, 14b are output samples for the modulator arrangements in Figures
5,
11 respectively. Rows I, II, III in each case represent output from each of
the three
first order modulators 51, 52, 53 and 111, 112, 113 before input to the
respective
delay elements 54, 114. Row IV in each case represents output from the
combination stages 55, 115. A limit cycle in the output of the first modulator
stage
is indicated and causes fractional spurs.
Figure 15 shows a further preferred modulator 150 formed from the modulator 60
of Figure 6 by addition of a dither signal in the feedback process. The dither
signal
is preferably pre-filtered as mentioned in relation to Figure 4d to avoid a
noise floor.
Only a finite number of feedback signals are generated by the logic stage 63
and a
new stage 153 is readily formed by combining the dither signal d as an
additional
input. As before a group of t msbs is received from adder 61. An m bit output
from
the logic stage is selected from a set of predetermined coefficients according
to the
values S and D. The output of adder 64 now a quantisation error signal with a
dither signal to reduce the likelihood that a limit cycle will develop for a
particular
control word X. A cascade of modulators such as shown in Figure 11 may include
one or more dithered stages.
Figure 16 is a table indicating a possible range of coefficients for the
feedback logic
stage 153 when the modulator 150 implements a two level quantiser function. F
is
the coefficient value which would be passed to adder 64 in the absence of
dither.
L represents the value added in response to the dither signal d. The value of
L is
limited to avoid overload and an increase in background noise.
A modulator according to the present invention can be used in a variety of
electronic
systems other than frequency synthesisers. In digital-to-analog conversion for
example. It is possible to use the modulator in various cascade arrangements
such
CA 02370254 2001-10-12
WO 00/62428 PCT/NZ00/00054
-13-
as shown in Figure 11. It is also possible to form a higher order system by
nesting
a number of modulators and providing suitable feedback.