Quadrature Sampling Scheme With Improved Image Rejection For Complex-IF Receivers

A Quadrature Sampling Scheme with Improved
Image Rejection for Complex-IF Receivers

Kong-pang Pun, Jose' E. Franca and C. Azeredo-Leme
Instituto Superior Tknico, AV. Rovisco Pais 1, 1096 Lisbon, Portugal
E-mail: kppun,franca, [email protected]
Abstmct- In complex-IF receivers, I/Q imbalances will
limit their image rejection performance. In this paper, a
quadrature sampling scheme combining the function of IFto-baseband conversion is proposed t o transform the image
problem caused by I/Q imbalances in this stage to the less
important self-image problem. With this scheme, signal is
sampled at IF stage but digitized at baseband. DC and other
low-frequency noises are circumvented while baseband AX
modulators can be used. Circuit simulations show that the
proposed method removes completely the image interference due t o I/Q imbalances existed after the IF input.
+IF1
-IF1
UF.
I. INTRODUCTION
+q
+IFq
Digital radio receivers with complex intermediate frequency (IF) stage have on-chip image rejection capability and attracted great attention recently[l], (21, [3]. In
complex-IF receivers, the complex I F signal is usually directly digitized by two real bandpass A X analog-to-digital
converters (ADC) or one complex bandpass 4 E ADC.
Since non-zero I F is used, it avoids many pitfalls existed in
zero-IF receivers, like spurious local oscillator (LO) leakage, DC offsets caused by self-mixing of LO and interferes,
flicker noise, etc. However, complex-IF receivers suffer from
the well-known image problem. If there are gain or phase
imbalances between the in-phase (I) and quadrature (Q)
paths, interferers a t the image frequency will be folded to
the signal band. For example, gain mismatch of 1% will
introduce a n image of -40dBc. In general, a receiver must
reject image interferers by at least 60dB. Therefore high-Q
front-end image filters, usually off-chip, are still required to
fulfill this requirement.
Recently. a double quadrature sampling (DQS) scheme.
which performs complex I F signal sampling and IF-tobaseband conversion by discrete-time analog circuit techniques, was proposed [4]. It inherits all the advantages
of direct-IF digitizing receivers but uses baseband AX
converters rather than bandpass converters Again. this
method has limited image rejection performance due to
channel mismatches or phase eirors.
To improve the image rejection III complex-IF receivers. a
quadrature sampling technique independent on the channel
mismatches as well as phase errots is proposed in this paper. Similar to the previous method, the proposed method
samples the signal at I F stage and performs IF-to-baseband
conversion in the discrete-time analog domain, all realized
in the sampling units of baseband I and Q AC ADCs. The
T h i s work \\.as supported i n part I,?. the FCT of the hlinistry
of Science and Technology of Portugal under the grant PRAXIS
S S I / R D / I Y888/99.
Fig. 1. Double quadrature sampling approaches. (a-ti) In digital

domain; (c) In analog domain.
main contribution of the proposed method is the elimination of the image problem caused by channel mismatches
and phase errors. Any such errors existed after the IF input will not cause the image interferer to be folded t o the
desired signal band. Only the so-called self-image, which
is not critical since it origins from the desired signal itself,
can arise from I/Q imbalances.
The rest of this paper is arranged as follows. In section 11, the principles of quadrature sampling in complex
I F receivers and its realizations by digital and analog circuits are presented. Mismatch and A/D conversion issues
are discussed. Section I11 presents the proposed method.
Section I\- shows circuit simulation results, and section 1concludes the paper.
11. QUADRATURE
S A h l P L I N G I N COMPLEX
IF
RECEII'ERS
Sampling a complex signal in quadrature form is referred
as double quadrature sampling. Different DQS approaches
are shown in Fig. 1. All of them sample the I and Q inputs
directly a t IF stage. Based on the different approaches on
IF-to-baseband conversion, they can be categorized into
two classes as described as follows.
A . IF-Sampling IF-Dagitiziny
The first kind of DQS methods is t o sample and digitize
the complex input signals a t IF stage, and then perform
the complex IF-to-baseband conversion in digital domain.
Fig. l(a-b) shows two varia.tions of this method. Fig. l ( a )
uses two .4DCs with sampling frequency at, least, two times
1-45
0-7803-6685-9101/$10.0002001 IEEE
Fs
(c)
the input signal bandwidth. Bandpass noise shaping must

be employed if oversampled ADCs are used.
Fig. l(b) is a variation of Fig. l(a). It uses a complex
bandpass 4 E modulator to digitize complex I F signals instead of two real ones. A complex bandpass 4 E modulator
bas notches of its noise transfer function mainly in one side
of the frequency axis. [a]. Normally its complexity and
power consumption is less than that of two real ones for
the same degree of noise shaping.
It is a common practice to set the sampling frequency fs
of the -4,DC four times of the input center frequency so that
the I and Q digital LO sequence are simply 1 , 0 , - 1 , O , . . .
and 0.1,0, - I , . . . respectively.
units of AX ADCs. h'loreover, the function of sub-sampling

can be also realized by this circuit if wanted.
+IF1
A.1 Mismatch Effects
-IFI
The DQS methods shown in Fig. 1 (a-b) have two analog

paths and therefore have the the problem of channel mismatch. Let us consider the Fig. l(a) first. Denote the I and
Q input spectrum as S I ( J Wand
) X Q ( ~ Wrespectively,
)
and
let the gain of I and Q ADCs be G, and G, respectively.
I t can be easily proven that, with the mismatches, the IF
signal becomes
where S ( j w ) = X ~ ( j w +
) j S ~ ( j w ) The
.
first and second
term of (1) represents the nominal input and a mirrored
version of it. Therefore, a. portion of the image interferer
is folded to the wanted signal band, and vice versa. Define
the image rejection ratio (IRR.) as the rat,io of the power
gain of the desired signal S ( j w ) over the power gain of the
image signal S * ( j w ) . From ( l ) ,we have I R R = ~ , ? ?
Let G;, G, = G f AG: where G is the nominal gain value
and AG is t,he deviation. We have
For the DQS method shown in Fig. l ( b ) , the mismatch

effect, is the same [SI.
B. IF-Sunipling Busebu7id-Digitizirig
The second kind of DQS methods is to sample the conipies input signals a t I F stage, perform the coniples IF-tobaseband conversion in discrete -time a.nalog domain, then
the .4/D conversions in baseband, as shown in Fig. l ( c ) .
This method is obtained a.ctually by moving forward the
complex IF-to-baseband conversion before the A/D conversion. Since signals at, t,he input, of ADCs lies in baseband.
liasehand AC ADCs can be used.
In Fig. l ( c ) , I F inputs +IF;,+IF,,-IF;, and -IF,I arc
sampled by I channel during clock phase 1, 2, 3 arid 4
respectively; +IFq,+IFj,-IFq, and -IF; are sampled bj.
Q channel during clock phase 1, 2, 3 and 4 respectively.
The I and Q outputs are equivalent to [ X l c o s ( n n / 2 )-YQ sin(n7i/2)] and [XQcos(7~7r/2)+ -Til sin(nn/2)], 71 =
0, 1,2 , . . ..
A circuit, realization of the special sampling technique is
shown in Fig. 2 [4]. It, is actually realized by the sampling
J......
J . J .
Fig. 2. Switched-capacitor double quadrature sampling circuit realized in the sampling units of two baseband AX .4DCs.
B.l Mismatch Effects

The DQS method shown in Fig. l ( c ) has four analog
paths. To analyze the mismatch effect, let us model the
sampling process as shown in Fig. 3, where t h e PI and PQ
are I and Q sampling signals. Denote the gain in four paths
as G,, , G,, , G,, and G,, as shown in Fig. 3. It; can be easily
proven that the output J j j w ) is
where @ denotes convolution, * represents complex conjugate and
~~~~.
and X ( j w ) = X i ( j w )
j . Y ~ ( j w ) , P ( j w ) = P~(jw)
j P ~ ( j w ) The
.
output consists of two parts: input X ( j w )
sampled by P,,,, and input conjugate X * ( j w ) sampled by
I'd,,,.
The spectra of P,,, Pdzff
and output l ' ( j w ) are
shown in Fig. 4.
Fig. 3. The mismatch model of the DQS method of Fig. l ( c ) .
The baseband output consists of a wanted signal and

three unwant,ed interferers, namely, 7 b l : n2 and n3 as shown
in Fig. 4. Among these interferers, 113 is the image of the
1-46
(=)
....
;k.;. ..."?a;P..,
+
0 w 4
Fig. 5 . The proposed mismatch- and phase-error-free double quadrature sampling circuit.
0 w4
Fig. 4. Spectra developed in analog double quadrature sampling

scheme. (a) input and its complex conjugate; (b) common-mode
and differential sampling signal, where A , = (Gtt G,,
G,,
= (Gat - G q , + G q Gqq)/4,A2 = (Gtt+Gqt -GTq-Gqq)j4,
G q 9 ) / 4and A4 = (G,, - G,, - G,,
G,,)/4; and (c) output.
are shared in all the sampling phases for output I and Q

respectively, we have Gii = G,, and Gig = G,i. So the
P,, and P & f f of (4) and (5) becomes:
wanted signal itself, it affects only the signal t o noise ratio

and is not harmful. Moreover, this kind of interferers can
be corrected by some digital correction methods [GI, [7],
[8]. Only n1 and n2 are problematic since they are origins
from another radio channel which may much stronger than
the wanted. Define the image rejection ratio as the ratio
of the power gain of the desired signal over the power gain
of images n1 and n'. From equation (3), we have
Referring t o Fig. 4, A2 and A3 become zero. The image

terms n l and n2 are eliminated. Only the self-image component n3 remains. As mentioned in the previous section,
the self-image problem is not severe and can be compensated in digital domain.
A . Second Order Eflects

Let Gii, G,i, Gi,, G,, = G fAG, where G is the nominal
gain value and AG is the deviation, we obtain
IRR
1
"
+ 2(24G)? - ( % ) 2 .
(4G)
2(2AG)'
(7)
For example, for

= 0.1%. we have I R R = 60dB.
It is interesting to know. from equation ( 2 ) and (7).
that both the IF-sampling IF-digitizing and IF-sampling
haseband-digitizing DQS methods have same image rejection ratio foi the same channel mismatch conditions, although the later has more aiialog paths.
111. THEPROPOSED
SAMPLING
SCHEME
Fortunately, a DQS circuit free of channel mismatch and
non-even phase problem is found. The circuit is shown in
Fig. 5 . The period of sampling clock 1: 2 : 3 and 4 is halved
in this circuit. Besides, the falling edges of these clocks
I. arrangement has
are a bit later than that of clock *-This
two purposes. First. it, reduces the clock-feedthrough arid
charge injection errors. Second, and more importantly, the
sampling instant is now det,ermined by the falling edge of
clock -4 only. Therefore, the effect of relative t,iniing errors
among clock signal 1, 2 , 3 and 4 is avoided.
On the other hand, the sampling capacitor C1 and Cz
are shared in all thr four phases for channel I and Q respectively. The mismatch exists het,ween Cl and C. , but
it produces self-image only as explained in the following
paragraphs.
The mismatch model shoxn in Fig. 3 can be still used

for the new circuit. But. since the capacitor C1 and C2
First, let us consider ON-resistance mismatches among

sampling switches, which will lead to different settling
time in the four sampling phases. Assume that the ONresistances of all sampling switches are constant. A relative
error in the ON-resistance, denoted as ARIR, will give a
relative error in the voltage charged t o the sampling capacitor (C, or 6 ' 2 ) . denoted as AlL/LL. I t can be easily shown
that this error is given by
ill
1;
e - m
T
AR
1- e m , ) ,
(
for
>> 1 and
<< 1,
(10)
where T = l / f s , R is the nominal ON-resistance, C is

the capacitance of the sampling capacitor. The effect of
AVJI, is similar to the effect of channel mismatch. Substituting A V J l ~ for AG/G in ( 7), we obtain the IRR as a
function of the ON-resistance mismatch. Obviously, to reduce ill :./I>, one must decrease the switch ON-resistance.
For example, if
= 1%,we have IRR of 69 dB and 84 dB
for RC = T/10 and RC = TI14 respectively. When the
RC time constant is no greater than one-seventh of a half
2 7), the IRR is larger than 84 dB
sampling period (&
and the OK-resistance mismatch effect can be ignored.
Consider the charge injection now. The charge injection
from input switches is eliminated by the special clocking
scheme. Charge injections from other switches are signalindependent, constant in all the four sampling phases and
can be cancelled by using fully differential circuit topology. Therefore, this non-ideality will not affect the image
rejection performance of the proposed circuit.
Last, other kinds of circuit non-idealities, like different
k T I C noise in I and Q paths, finite gain and bandwidth of
amplifiers. are signal-independent and equal in all the four
1-47
sampling phases. They will not affect the image rejection

performance of the proposed circuit.
IV. S I ~ ~ U L A TRESULTS
ION
Circuit simulations of the existed and the proposed
DQS together with ADCs were conducted by using SWITCAP [9]. The ADCs adopted here are one-bit, third order
cascaded (1-1-1) AX modulators [lo] with lowpass noise
shaping. The switched-capacitor circuit diagram of the first
modulator is shown in Fig. 6. Note that only single-ended
circuit is shown for clarity. To carry out those quadrature
sampling schemes, the sampling unit of Fig. 6 should be
replaced by circuits shown in Fig. 2 and Fig. 5. Note that
this modulator is selected arbitrarily. ilny other baseband
modulators are applicable here as long as they can be realized in discrete-time circuit form.
m p l l n g Unll
iffi
UL - r L n
Fig. 7. SWITCAP simulation results: (a) input spectrum; output

spectrum of (b) the existed DQS circuit; (c) the proposed DQS
circuit; Fixed channel mismatch of 1% is assumed
-vnt
Fig. 6. Circuit diagram of the first modulator in a 3rd ordw AX

ADC used in the simulations.
In the simulations, two complex input tones are applied

as shown in Fig. 7(a). The desired signal and image interferer lie at -(0.25 + 2 - 5 ) f s and (0.25 2 Y 6 ) f s respect,ively.
The output responses are shown in Fig. 7(b-c), where fixed
channel mismatch of 1%is assumed.
scheme transforms the image problem caused by I / Q mismatches to the less-iniporta,nt self-image problem. Circuit
simulations proved that, complete image rejection can be
achieved in this IF-to-baseband stage in the presence of
I/Q mismatches and phase errors. Only self-image arises
from I / Q mismatches.
REFERENCES
. I . Crols ancl kl. Steyaert, ;\ single-chip 900 h l H z CMOS receiver front-end with a high performance low-IF topology, IEEE
. I . Solid-State Circuits, vol. 30, no. 12, pp. 1483-1492: Dec. 199.5.
S.A. .Jantzi c t al., Complex handpass EA conveitrr for digital
radio. fi.oc. I E E E ISOlzYLl4: vol. 5. pp. 45:3-43;> iblay-,June
For the existed DQS circuit, Fig. 7(b) shows that the
desired signal is down-converted to baseband a t

and the image interferer is shifted to around f , 5 / 2 . Due to
the channel mismatch, a portion of the image interferer is
folded to t,he baseband at f 2 - f S , as indicated by II and
12. Besides, a. self-image, 13, also appears in the baseband
at + 2 - f S . The I1and 1.1are suppressed by 46dB in magnitude, which means the image rejection ratio of 40dB arid
agrees with equation ( 2 ) .
For the proposed DQS circuit, Fig. 7(c) shows that, the
desired signal is also down-converted to baseband but there
are no image components at f 2 p 6 f , . Only the self-image
13 at + 2 - 5 f s , with a magnitude of -40dBc, arises from
the fixed 1% channel mismatch. This result corifirnis our
prediction.
V. CONCLUSIONS
A quadrature sampling scheme has been proposed for
complex I F receivers to improve their image rejection performa.nce. In this approach, the ideal I / Q mixer is moved
from the digital signal processing unit t o the analog sampling unit, so that most advantages of the direct-IF digitizing approach, including the immunity to DC offset error,
are preserved while lowpass AX ADCs instead of bandpass ones are required. More importantly, the proposed
1994.
.J.C;. Kudell, . J . J . Ou? et al.: ..-II.9GHz \vide-h~~.iid

IF double
conversioii CMOS receiver for cordless telephone application,
IEEE .J. Sulid-State Circuits. vol. 32. pp. 2071-2088, Dec 1997.
C . Azeretlo Leme, Ricardo Reis, aiitl Eduardo \:irgas, Wideband sub-sampling A / D conversioii with image i.e,jection, in
I E E E Workshop on Wireless Comiriunicatioii Clir~;iiit.sand Systems, Lucerne, Switzerland: .lune 1998.
A . Swaminatliari, .4single-IF receiver architecture using a coniplex sigma-delt,a motlulat.or, h1.S. thesis. ChrletGxi Universit,y,
Ottawa, C:aiiada, 1997.
F.E. Churchill, G.W. Ogar, and B..J. Thompson. The correction
of I and Q errors in a coherent processor, IEEE Transactions
on. Aerospuce and Electronic Systems, vol. AES-17, no. 1, pp.
1.31-137, .Jan 1981.
M.D. Kulkarni and Alexander B. Iiostinski, A simple formula
for monitoring quadrature phase error with arbit,rary signals.
IEEE Trail. Geoscience and remote sensing, vol. 3 3 , no. 3, pp.
799-802. klay 1995.
It. A. Green, K. Antierson-Sprech~,r,and J. W. Pierre, Quadrature receiver mismatch calibratiou. IEEE Trun. Signal Processing, vol. 47, no, 11, pp. 3130-3133, Nov. 1999.
I<. Suvama and S.C. Faria. User% M(inria1 f o r S W l T C A P 2 uersi071 1.1, Columbia University, 1992.
[lo] Y. Matsuya, K . Uchimura, A. Iwata, et, al., A 16-bit oversampling A-to-11 conversion t,echnology using triple-integration
noise shaping; IEEE .I. Solid-State Circuits, vol. 22, pp. 921929, Dec. 1987.
1-48

Quadrature Sampling Scheme With Improved Image Rejection For Complex-IF Receivers

Uploaded by

Copyright:

Available Formats

Quadrature Sampling Scheme With Improved Image Rejection For Complex-IF Receivers

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quadrature Sampling Scheme With Improved Image Rejection For Complex-IF Receivers

Uploaded by

Copyright:

Available Formats

A Quadrature Sampling Scheme with Improved

Image Rejection for Complex-IF Receivers

Fig. 1. Double quadrature sampling approaches. (a-ti) In digital

the input signal bandwidth. Bandpass noise shaping must

units of AX ADCs. h'loreover, the function of sub-sampling

A.1 Mismatch Effects

The DQS methods shown in Fig. 1 (a-b) have two analog

For the DQS method shown in Fig. l ( b ) , the mismatch

B.l Mismatch Effects

where @ denotes convolution, * represents complex conjugate and

Fig. 3. The mismatch model of the DQS method of Fig. l ( c ) .

The baseband output consists of a wanted signal and

Fig. 4. Spectra developed in analog double quadrature sampling

are shared in all the sampling phases for output I and Q

wanted signal itself, it affects only the signal t o noise ratio

Referring t o Fig. 4, A2 and A3 become zero. The image

A . Second Order Eflects

For example, for

The mismatch model shoxn in Fig. 3 can be still used

First, let us consider ON-resistance mismatches among

where T = l / f s , R is the nominal ON-resistance, C is

sampling phases. They will not affect the image rejection

Fig. 7. SWITCAP simulation results: (a) input spectrum; output

Fig. 6. Circuit diagram of the first modulator in a 3rd ordw AX

In the simulations, two complex input tones are applied

desired signal is down-converted to baseband a t

.J.C;. Kudell, . J . J . Ou? et al.: ..-II.9GHz \vide-h~~.iid

You might also like