ACE9030
The value of SF can be found from any channel but to get
a quick estimate the highest frequency can be considered as
there is a fixed upper limit on CN of 256 so:
For a typical AMPS cellphone the fVCO(max) for 45 MHz
I.F. is 938·97 MHz, fCRYSTAL is 14·85 MHz, MOD is 8 and
CN(max)canbeassumedtobechosenaround200,givingIco
= 0·198 x Ibo.
CN(max)
SF =
NTOT(max)
Fractional-N Mode with Speed-Up
Putting suggested typical values into equation (7);
NTOT(max) = 10000, CN(max) = 250, and MOD = 8, and then
assuming the current flows for the whole comparison period
the current to be multiplied by ACC is Ibo / 320. The typical Ibo
is only 1 µA so this is a very small current of around 3 nA and
wouldbetoosmalltocontrolaccuratelyandcertainlytoosmall
for production testing.
The error signal to be cancelled is a narrow pulse at the
comparison frequency so the best cancellation of the whole
spectrum of the error is also a narrow pulse. It is not practical
to generate a variable width pulse to match the error pulse but
a fixed width variable amplitude pulse is possible and it can be
timed to approximately coincide with the error pulse to give
good cancellation.
The compensation current amplitude is also increased by
gating it with a small time window and in ACE9030 the gate is
set to two cycles of the reference clock which straddle the
active edge of the comparison frequency signal to the phase
comparator. The total reference division from reference clock
to comparison frequency is the programmable divider set by
NR in Word D multiplied by 1, 2, 4, or 8 as selected by the SM
bits also in Word D and this total may be called RMAIN, so for the
compensation current the scaling is RMAIN / 2.
When Speed-up is active the main proportional charge
pumps are run at an increased current and the integral charge
pumps are switched on to move the loop filter voltage faster.
The phase errors due to Fractional-N mode will be the same
as normal once the loop is locked so the compensation pulses
must be increased to match the proportional and integral
charge pump currents in order to allow a smooth change over
to normal mode at the end of Speed-up time. The same 2L + 1
and K coefficients as used for the proportional and integral
chargepumpcurrentsareusedonthecompensationcurrents
so from equation (8):
Normal Mode:
Proportional Compensation Current:
Icomp(0) = ACC x Ico
Integral Compensation Current:
none = off
Speed-up Mode:
Proportional Compensation Current:
Icomp(1) = 2L + 1 x ACC x Ico
Integral Compensation Current:
Icomp(2) = K x 2L + 1 x ACC x Ico
The charge needed is still as in equation (7) but the
current can be defined as:
Required Accuracy of Compensation
With the compensation scheme used in ACE9030 it is not
possible to get perfect cancellation of the loop disturbance by
the Fractional-N system due to the mis-match of the pulse
shapes leaving some high frequency terms, but if the areas
are matched there will be complete removal of the low fre-
quency components and the loop filter can be assumed able
to remove higher frequencies.
Typical timing waveforms for the phase error and its
compensation are shown in figure 28 for a loop operating in
1/8’s mode (hence MOD = 8), with a VCO at 1 GHz, and a
comparison frequency of 100 kHz (hence NTOT = 10,000 and
fCOMP period = 10 µs) so that each phase error can be found
from equation (4) as:
Icomp (0) = ACC x Ico .....(8)
where, in ACE9030, the compensation reference current Ico
is set by an external resistor RSC such that:
IRSC
320
Ico =
but this Ico must be chosen to cancel the error charge in
equation (7), and the scaling effect of 2 reference cycles in
RMAIN has been derived above, giving:
(ACC x 10 µs) / (10,000 x 8) = ACC x 0·125 ns.
If the reference is a 12·8 MHz crystal, it gives a correction
SF
MOD
RMAIN
2
Ico =
x
x Ibo
pulse
duration,
two
reference
cycles,
of
2/(12·8 MHz) = 156 ns.
If the charge pump current of 250 µA is set by a CN value
of 250 the reference current Ibo from equation (6) is 1 µA and
thecompensationstepcurrent Icocanbefoundfromequation
(10) as 0·2 x Ibo = 0·2 µA.
Areas of the error and the compensation pulses, equa-
tions (4) and (8) must match to get good low frequency
cancellation. AlthoughshownasaverynarrowpulseonØDOWN
the phase error will often appear as a change in size of the
pulses on either ØDOWN or ØUP which occur to maintain lock.
The following calculations would then apply to the changes
and give the same final result.
IftherewasnocompensationtheØDOWN pulseswouldgive
sidebands at a level set by the loop filter capacitor values and
the VCO gain.
In a typical system the filter proportional capacitor can be
6·8 nF and the VCO could cover 30 MHz in 3 V, giving
10 MHz/V. Assuming for the moment that all error pulses are
the same at the level of a mid-range ACC value, say 4, and do
then removing SF to help evaluate the values needed:
CN (max)
RMAIN
2
Ico =
x
x Ibo ....(9)
MOD x NTOT(max)
this can then be further processed by replacing RMAIN and
NTOT(max) by the frequency ratios:
fvco (max)
fCOMP
fCRYSTAL
fCOMP
RMAIN =
and NTOT (max) =
then when substituting these into equation (9) the fCOMP terms
cancel leaving:
CN (max)
fCRYSTAL
Ico =
x
x Ibo ....(10)
2
MOD x fVOC (max)
32
MITEL [ MITEL NETWORKS CORPORATION ]
