Ri (il)
600Mi
16x2.5 0.18 1.67C\ (PF)
16.5Msw
24x2.5 0.18 0.84Table 6.2: SBM
design
values.6.4.2
LO Phase Shifter
There are several ways to
implement
a 90°phase
shifter.Here,
beforegoing
into the details of the chosenimplementation,
webriefly
discussthe four
predominant
methods used withinintegrated-circuits:
• RC — CR Network: A 90°
phase
shifter can be realized withLOo-
Figure
6.7: RC—CRphase shifter.
an RC
low-pass
and a CRhigh-pass
filter as shown inFig.
6.7.Provided the resistors and the
capacitors
arematched,
thephase
difference is
independent
fromfrequency.
If the two filters are6.4.
Image-Reject
Mixer 55loaded
by
matchedloads,
thephase
difference results also inde¬pendent
from the load. On the otherhand,
theamplitude
is thesame
only
at the -3 dBfrequency
which is setby
an RCproduct, and,
due to processvariations,
in a worst-casescenario canvary as much as ±40%. To restoreequal amplitudes limiting ampli¬
fiers
following
thephase-shifter
are needed. To be effective suchlimiting amplifiers requires
aquite high
powerconsumption.
A furtherdisadvantage
of this type ofphase-shifter
is the need foranLO buffer. The
lossy input impedance
of the networkcannot bedirectly
drivenby
an LC oscillator.Sequence Asymmetric Polyphase
Network: A 90°phase
shifterwhich also introduces
only
smallamplitude
errors can beimple¬
mented with the
help
of sequenceasymmetric polyphase
net¬works,
often calledpolyphase
filters. A two stagepolyphase phase-shifter
is shown inFig.
6.8. The RC time constant of eachLO
Q
Figure
6.8:Polyphase phase shifter.
sectiondetermines the
frequency
atwhich theamplitude
and thephase
errors are zero.By cascading
several sectionsa broadbandphase-shifter
can be realized. Due to thecyclical
connection ofthecomponents, a
polyphase phase-shifter
isless sensitivetocom¬ponentmismatch than an RC—CR one. A
disadvantage
of thisstructure is the attenuation introduced
by
the network. A two stagephase-shifter
which has beensuccessfully
used in a proto-56
Chapter
6.Design
of the Receivertype
during
thisproject [\ '»]
shows an attenuation of 3.2 dB and isthuscomparable
tothe attenuation introducedby
anRC—CRphase-shifter. Also, similarly
to an RC —CRphase-shifter,
thelow and
lossy input impedance
of the network dictates theuseofan LO buffer
amplifier.
•
Coupled
VCOs: A further method which can be used to gener¬ate
quadrature
/ andQ
LOsignals
isby coupling
two oscillators with the same nominal oscillationfrequency
asproposed
in[', ].
If the oscillators are
coupled
tooloosely,
the oscillators do notsynchronize.
If thecoupling
is too strong thephase-noise
de¬grades.
Therefore thequadrature
oscillatorperforms
asexpected only
over a limited range of thecoupling
coefficient[
*>].
Sincethis solution does not
require
bufferamplifiers,
it shows aslight advantage
intermsofpower. Thisadvantage
ishoweverpartially
lost dueto the need fortwo oscillators.
•
Digital
Divider: Adigital divide-by-two frequency
divider consti¬tuted
by
two latches clocked on theopposite edges
of the clock generatestwosignals
with aphase
difference of 90° whichcanbe used as / andQ
LOsignals (Fig.
69).
This kind ofphase-shifter
C•>
Q
o
r LD Q
D-latch
]CLKQ
D Q
D-latch
]CLKQ
Lo r
r°
(2*fL0)
O-L-
Figure
6.9:Digital divide-by-two phase shifter.
is very wideband
(ideally
from DC to anupper limit determinedby
thelatches),
verycompact, and veryprecise.
The main draw¬back is the need for an oscillator
running
at twice therequired frequency.
Thisdisadvantage
in a transceivercan be turned intoan
advantage by considering that,
due to the differentfrequen¬
cies,
no interaction between the output of the PA and the LO6.4.
Image-Reject
Mixer 57signal
can occur. At GHzfrequencies
the powerconsumption
offrequency
dividers becomesrelatively high, but, considering
the buffers or
limiting amplifiers required by
othersolutions,
thepower
consumption penalty
is not verysevere.Due toitsmany
advantages
we have decidedtoimplement
adigital divide-by-two frequency
dividerphase-shifter.
The main factorlimiting
the achievable
quadrature
accuracyusing
this kind ofphase-shifter
isthe even-harmonics content of the VCO
signal driving
thecircuit,
as even-harmonics cause theduty cycle
of thesignal
to deviate from the ideal 50% value. Weanalyze
thephase
errorgenerated by
the second- harmonicalone,
sincehigher
even-harmonics arealways significantly
lower than the secondone.
Fig.
6.10shows thephase-error
agenerated by
thesignal
A[sia(ujt)
A-Ksia(2ujt A-4>)} (6.16) driving
an ideal master-slaveD-flip-flop
as a function of the relativeamplitude
of the second-harmonic component with respect to the fun¬damental
K,
for threephases </>.
Thephase-error
a isamplitude
andphase dependent:
if the second harmonic component isin-phase
withthe fundamental
(</>
=0),
no error isproduced;
the maximum error is causedby
aphase
difference</>
of 90°. Asusually
it is notpossible
tocontrol the relative
phase </>,
theonly
meantominimizethephase-error
ais to
keep
the second-harmonic and allhigher
even-harmoniccompo¬nents as low as
possible.
For this reason it is veryimportant
to use a VCO with a differentialtopology.
The
implemented divide-by-two frequency
divider(Fig. 6.1])
hasbeen
designed
inenhancement sourcecoupled logic (ESCL)
with resis¬tive loads. The circuit is
composed by
acouple
of latches clocked inanti-phase. Having
afully
differentialstructure ESCL gatesinject
very little noise into the powersupply
and thesubstrate,
and are immuneto common-mode noise.
Furthermore, working
with low levelsignals they
aremuchmorepowerefficient than otherlogic
families. The static powerconsumption
of thislogic family
is ofno concern here since the divider isalways working
at fullspeed.
The
predominant
RC timeconstantlimiting
thespeed
of the circuit is constitutedby
the load resistors and the totalcapacitance loading
the output nodes.
Thus, knowing
the loadcapacitance,
therequired
58
Chapter
6.Design
of the Receiverrelative
amplitude
K[dB]
Figure
6.10: Relativephase-error
between the L andQ signals
as afunction of
the relativeamplitude of
the second-harmonic withrespectof
the