Beruflich Dokumente
Kultur Dokumente
3, MARCH 2019
Abstract— This paper demonstrates a tunable synthetic to meet compact form factor requirements, there has con-
fourth-order bandpass filter (BPF) at microwave frequencies. sequently been a great drive to explore on-chip filtering
Two parallel second-order Q-enhanced LC BPFs responses are techniques, including N-path, Q-enhanced LC, active LC,
added with the out of phase to synthesize a fourth-order BPF
response. The filter is implemented in a 130-nm SiGe BiCMOS and discrete-time (DT) RF filtering. N-path filtering [1]–[20]
technology with a core die area of 0.53 × 0.7 mm2 . The filter has emerged as a promising candidate in terms of performance
center frequency can be tuned from 4 to 8 GHz (C-band). The for on-chip integration but has historically been limited to fre-
filter also achieves a wide 3-dB fractional bandwidth (BW) tuning quencies ∼3 GHz and below. N-path filtering requires gener-
range of 2%–25%, with a passband ripple of less than 0.5 dB. The ating accurate multiphase clock signals. Such clock generation
corresponding normalized dynamic range (DR) is 151–166 dB·Hz
owing to a switched varactor control scheme to realize a large circuits require excessive dynamic power consumption and
effective tuning range with high linearity. Using the parallel become difficult to realize accurately beyond low microwave
synthesis approach, the filter can maintain the DR of a second- frequencies. In addition, DT RF signal filtering approaches
order BPF while achieving a fourth-order frequency selectivity, have been reported for such receivers [20]–[22], [40], [41].
which is favorable compared to cascading resonators. On the These filters also require differential or multiphase clocks and
lower side of the band, the filter achieves more than 65 dB of
ultimate rejection. On the upper side, the rejection is more than hence have similar challenges as N-path filters.
52 dB. The filter also employs a variable transconductor for More recently, N-path filters at microwave frequencies of
noise-linearity tradeoff flexibility. The power consumption of the 6 and 12 GHz have been reported [14], [15]. These filters
filter is 112–125 mW over the above fractional BW tuning range have a very large frequency tuning range accomplished by
at the target C-band. local oscillator (LO) (clock) tuning; however, with the increas-
Index Terms— Bandpass filter (BPF), bandwidth (BW) tuning, ing operating frequency, the out-of-band (OOB) rejection
frequency tuning, Q-enhanced LC filter, varactor, wideband. decreases progressively to 20 dB or worse. This is due to
I. I NTRODUCTION the filtering dependence on the clock’s duty cycle which gets
distorted at a higher frequency since the rise and fall times are
Fig. 3. (a) Synthesizing a fourth-order bandpass response by subtracting two identical second-order resonators each tuned at a slightly different resonance
frequency. (b) Frequency spacing between resonators and BW of each resonator versus in-band ripple magnitude for a total BW of 300 MHz. (c) Spacing
between resonators and total BPF BW versus in-band ripple magnitude for a resonator BW of 200 MHz.
above ωo2 , the two waveforms will be subtracted. In between of ∼210 MHz between the two resonators, i.e., the resonators
the two resonance frequencies, the phases are opposite, and are centered at f o1 ≈ 5.79 GHz and fo2 ≈ 6.21 GHz with
therefore, the waveforms will be added together, forming Q 1 ≈ 57, and Q 2 ≈ 61, respectively. These values are
a passband. Mathematically, the transfer function of the well matched with circuit simulations. For a given resonator
subtraction can be written as BW, e.g., 200 MHz in this case, the maximum achievable
ω01 ω02 fourth-order BW and the frequency spacing between the two
Q1 s Q2 s
H (s) = 2 ω01 − ω (1) resonators are plotted in Fig. 3(c) against passband ripple. For
s + Q 1 s + ω01
2 s 2 + Q022 s + ω02
2
a 3-dB ripple, the maximum fourth-order BW is 613 MHz at
where Q 1 and Q 2 are the quality factors of individual second- 6 GHz center frequency, which is 3.1 times the BW of the
order BPFs. If both BPFs have equal BW, i.e., ω01 /Q 1 = individual resonators. It is possible to achieve a very large
ω02 /Q 2 , then (1) becomes BW by lowering Q of each resonator to around 4–5 while
maintaining a fourth-order stopband roll-off, as discussed in
ω01 ω02 ωQ021 − ωQ012 s Section III.
H (s) = . (2)
s 2 + ωQ011 s + ω01
2 s 2 + ωQ022 s + ω02
2
B. Bandpass Filter Noise
In traditional filter synthesis [23], the design specification
A system-level noise analysis is performed for the fourth-
(ripple and BW) of the passband response determines Q and
order BPF. The block diagram of a synthetic fourth-order
resonant frequency of each LC tank. A similar process can
filter consists of two second-order filters A and B is shown
be followed in the fourth-order synthetic filter design. A set
in Fig. 4. The noises of the LC tank, the negative resistance,
of simple and approximate equations are derived from (2) in
and Q-tuning element are assumed to be dominant. The
terms of passband ripple (Mr in dB), center frequency (ωc ),
second-order BPF has noise sources from RC1,2 , used for
and 3-dB gain BW (ω3 dB ). The resonance frequencies ωo1 ,
Q control, tank losses R P1,2 , and the loss compensation
ωo2 , and Q of each resonator can then be set by the following
negative resistance R N1,2 = −1/G m1,2 where G m1,2 could be
equations. The derivations of these equations are provided in
a cross-coupled transconductance in typical Q-enhanced LC
Appendix A
filters. As can be seen in Section III-C, the noise contribution
ωo1 ωo2 ∼ ω3 dB from the unity gain buffers in Fig. 4 will be negligible since
= = √ (3)
Q1 Q2 10 M r /20 + 10 r /10 − 1 +
M 2−1 their output resistance will be far smaller than RC1,2 in actual
and transistor-level circuit implementation. The noise currents
from RC1,2 , R P1,2 , and R N1,2 are converted to voltages by
1 ωo1,2
ω ∼
= · 10 Mr /20 + 10 Mr /10 − 1 . (4) the total tank load Z Tank1,2 and the equivalent total noise
2 Q 1,2 power density (V2 /Hz) of each second-order BPF is given by
Consequently, each resonator center frequency can be set as
1 1
Vn,T 1,2 = 4kT
2 + + γEq G m1,2 |Z Tank1,2|2
ωo1 = ωc − ω and ωo2 = ωc +ω. (5) RC1,2 R P1,2
(6)
Fig. 3(b) shows the frequency spacing between the two
resonators and BW of each resonator when the total BPF filter where
BW ω3dB = 2π ×300 MHz is set in (3). For example, for a 5% ω01,2
γEq is the constant factor associated with the negative Similarly, the total noise power density will be given as in the
transconductance cell noise sources. In the vicinity of reso- following equation at the center of the passband (ωc ):
nance frequency, the tank impedance can be approximated to
2
Vn,T ω=ωc
RC1,2
Z Tank1,2 at ω = ω01,2 ± ω1,2 ∼
= R ω1,2
(8) = Vn,T 1 ω=ω
2 2
+ Vn,T
1+ j 2Q 1,2 RC1,2 2 ω=ω01 =ω02 −ω
P1,2 ω01,2 01+ω
RC1
where ω1,2 expresses a small offset frequency around ω01,2 = 4kT RC1 1 + 1+γEq
R
satisfying ω1,2 {ωo1 , ωo2 }. For simplification, let us P1
assume that RC1 = RC2 , R P1 = R P2 , and G m1 = G m2 . noise power of standalone 2nd −order BPF
⎡ ⎤
If the tank loss is completely compensated by the negative 2 −1 2 −1
RC1 RC2
resistance, e.g., G m1,2 = −1/R P1,2 , then the effective total × ⎣ 1+ K + 1+ K ⎦ . (11)
R P1 R P2
noise power density at ω01 will be given as
2
2
2
added nosie power factor ③
Vn,T ω=ω01
= Vn,T 1 ω=ω01 + Vn,T 2 ω=ω01 =ω02 −2ω
From (9)–(11), the added noise power factors, the noise
RC1
= 4kT RC1 1 + (1 + γEq ) penalty paid for the fourth-order BPF synthesis, are the
R P1
⎡ ⎤ function of RC1,2 /R P1,2 and K. The parameter K ranges from
2 −1
R 2ω 1.4 to 1.6 when 0.5–1 dB of passband ripple is allowed.
× ⎣1 + 1 + 2Q 2 ⎦
C2
R P2 ω02 In typical moderate-to-high-Q control greater than the intrinsic
LC tank Q, RC1,2 will be greater than the intrinsic LC tank
RC1 resistance R P1,2 . This makes the added noise power factors
= 4kT RC1 1 + (1 + γEq )
R
P1 ① and ② in (9) and (10) less than 1.1 if K = 1.5 is assumed,
noise power of standalone 2nd −order BPF
namely, less than 10% noise power increase at the edges of the
⎡ ⎤
2 −1 passband compared with the stand-alone second-order BPF.
R The added noise power factor ③ in (11) will be less than 1
× ⎣1 + 1 + 2 ⎦.
C2
K (9)
R P2 when K· RC1,2 > R P1,2 , resulting in a slightly better noise per-
formance than stand-alone second-order BPF at ωc . The reason
added nosie power factor ① for this better noise performance is obvious since the two
added noises from individual filters are perfectly uncorrelated,
In (9), 2ω/ω02 = K/Q 2 is applied where K = 10 Mr /20 + whereas signals are correlated at ωc . Note that as increasing
(10 Mr /10 − 1)1/2 in (4). Likewise, the effective total noise the ratio of RC1,2 /R P1,2 to obtain higher filter Q, the net noise
power density at ω02 is found to be as power from the individual second-order BPFs also increases,
2 2
2
degrading overall noise performance and exhibiting typical
Vn,T ω=ω02
= Vn,T 1 ω=ω02 =ω01 +2ω + Vn,T 2 ω=ω02 noise-selectivity tradeoff in the Q-enhanced LC filters.
RC2 In a low Q control smaller than the intrinsic LC tank
2
Vn,T ω=ω = 4kT RC2 1+ (1 + γEq )
02 R Q, RC1,2 will be smaller than R P1,2 and the added noise
P2 power factors ①, ②, and ③ in (9)–(11) will increase only
noise power of standalone 2nd −order BPF
⎡ ⎤ incrementally. Apparently, in extremely low Q case where
2 −1
RC1 K · R C1,2 R P1,2 , the two second-order BPFs could
× ⎣1 + 1 + 2 K ⎦. (10) degenerate to an all-pass filter at the frequency band of
R P1
interest. Then, the two BPF noise powers Vn,T 2 2
1 and Vn,T 2
added nosie power factor ② will be added together with the negligible noise shaping effect
772 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 54, NO. 3, MARCH 2019
Fig. 5. Conceptual description of nonlinear response of the synthetic fourth-order filter comprising two second-order filters
A and
B with two-tone analysis
for two cases: ① at the edge of fourth-order response and ② at the center of fourth-order response.
by the resonators. This is well expected in (9)–(10) where fundamental tones ( A ,fund ). The third-order input intercept
if K·RC1,2 R P1,2 the added noise power factors ①, ②, point (IIP3 ) of filter
A is found to be 10log(α· Pin ) [dB] where
and ③ approach to 2, manifesting non-coherent addition of Pin is the input power in a linear scale. Assuming the two
the two BPF noises at the fourth-order BPF output. It should fundamental tones (ω1 and ω2 ) and IMD3 tones (2ω1 −ω2 and
be noticed that as decreasing RC1,2 for lowering Q, the net 2ω2 − ω1 ) are close enough not to experience any meaningful
noise power from individual second-order BPFs also becomes phase delay difference between them, the two-tone response
smaller. Thus, even if the added noise power factors increase of filter
B can be given as (13)
the overall noise power will decrease, resulting in better noise
performance compared with the high-Q case. j ω1 t 1
A = e +e j ω2t + {e j (2ω1−ω2 )t + e j (2ω2−ω1 )t }
Intuitively each resonator attenuates noise power with a α
A A
second-order bandpass response in Fig. 4. Since the resonance ,fund ,IMD3
frequencies of resonators are staggered, the attenuated noise
δ · A ,fund
from filter A is either added directly to filter
B at ω = ω02
B = + δ3 · 1 e− j ϕ (12)
or further attenuated by the bandpass response of filter B α A ,IMD3
before being added at ω = ω02 . Thanks to this attenuated
bandpass noise shaping, the degradation of noise performance where
by the uncorrelated noise addition in the fourth-order BPF |Z Tank2|
is negligible. The net effect of the added noise at the edges δ= and ϕ = tan−1 (Z Tank2 ). (13)
RC2
of fourth-order BPF passband on the overall system noise
figure will be particularly trivial when the BPF is preceded In (13), the output of filter
B is modulated by a normalized
by a high gain amplifier stage such as LNA or transconductor tank impedance of δ · e− j ϕ . δ and ϕ are dependent on the
in typical RF front-end applications (see the circuit-level noise frequency offset from the resonance frequency of each filter.
analyses and simulation results in Section III-C). Note that the IMD3 tones are attenuated by a factor of δ 3
compared with the fundamental tone attenuation factor of δ.
C. Bandpass Filter Linearity The composite two-tone response (T ) at the output of the
fourth-order filter will be
Fig. 5 shows the nonlinear response of the fourth-order filter
when excited by two tones intuitively. The major sources of
T =
A +
B
nonlinearity are the varactor and the active loss compensation
− jϕ 1
negative resistance. Two cases are discussed for the two-tone = (1 + δ · e− j ϕ ) 3
A ,fund + (1 + δ · e ) .
location within the fourth-order filter band. Case ① is the α A ,IMD3
worst case scenario when the two tones are at the edge of (14)
fourth-order passband or at the center of the second-order
filter .
A For the nonlinear response of the individual second- From (8), the attenuation factor δ at 2ω offset from the center
order filters, shown on the top of each other, the filter
A will frequency of filter
B [see Fig. 3(a)] can be found as
give standard second-order filter response and its nonlinear
waveform 1 1
A is expressed as (12). In (12), the magnitude δ2 =
(1/α) of the third-order intermodulation distortion (IMD3) 2 = 2 . (15)
1 + 2Q 2 RRC2 2ω RC2
1 + 2 R P2 K
tones ( A ,IMD3 ) are normalized by the magnitude of the P2 ω02
AMIN et al.: INTEGRATED SYNTHETIC FOURTH-ORDER Q -ENHANCED BPF 773
Fig. 6. Simulated nonlinear response of the synthetic fourth-order filter comprising two second-order filters
A and
B with two-tone analysis for two cases:
① at the edge of fourth-order response and ② at the center of fourth-order response.
Therefore, IIP3 of the fourth-order filter at the edge of The attenuation factor δ at ω offset from the center
passband, IIP3 |T ,edge , is frequency of each filter [see Fig. 3(a)] is
1 1
IIP3 |T ,edge δ = 2 = 2 . (18)
RC1,2 ω
1 + 2Q 1,2 R P1,2 ω01,2
RC1,2
1 + R P1,2 K
(1 + δ2 cosϕ)2 + (δ2 sinϕ)2
= 10log α · Pin
(1 + δ2
3 cosϕ)2 + (δ 3 sinϕ)2
2 Therefore, IIP3 at the center of passband, IIP3 |T ,center , will be
⎛ ⎞
1
(1 + δ2 cosϕ)2 + (δ2 sinϕ)2 IIP3 |T ,center = IIP3 |filter
A + 10log ⎝
= IIP3 |filter ⎠. A + 20log δ . (19)
3 cosϕ 2 + δ 3 sinϕ 2
1 + δ2 2
The net IIP3 improvement is 20log(1/δ ) at the mid-passband.
(16) Fig. 6 illustrates a simplified simulation setup in Advanced
Design System (ADS) [42] to verify the linearity improvement
In typical Q control where K · R C2 > R P2 , δ could be much in the fourth-order BPF and simulation results for the cases
smaller than 1. Therefore, the resultant IIP3 is essentially the ① and ② in Fig. 5. In Fig. 6(a), the ideal voltage buffer
same as that of the second-order filter or slightly better. The is a voltage controlled voltage source (VCVS) component,
same is valid for the higher frequency edge of the fourth-order and the varactor from GlobalFoundries (GF) 0.13-μm SiGe
filter. BiCMOS process (GF8HP) library model (model name: ncap)
The best case, ② in Fig. 5, is when the two tones are at is used. The varactance ranges from 1.1 to 4.2 pF and is set
the center of the fourth-order filter passband. At the center of to 2.91 pF at filter A and 2.59 pF at filter B to resonate
passband, the two-filter outputs are added in phase and the the inductor L = 350 pH at around 5 GHz ( f o1 ) and
overall wave function of (14) will be modified to 5.28 GHz ( f o2 ), respectively, resulting in fc = 5.14 GHz.
In this setting, the varactor Q is 38.3 and 37, resulting in the
δ2 parasitic LC tank resistance of 420 and 430
in the filters
T =
A + A ,fund + α
B = 2δ e− j ϕ .
A ,IMD3
A and , B respectively. These resistances are compensated,
(17) respectively, by the negative resistances of RN1 and RN2 . Then,
774 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 54, NO. 3, MARCH 2019
Fig. 7. (a) Block diagram of the on-chip implemented differential synthetic fourth-order BPF. (b) Schematic detail of one path of the on-chip implemented
synthetic filter and subtractor.
filters experience slight different Q and nonlinearity because matched impedance or 1-k
high impedance. The emitter-
the varactors are biased slightly different bias points. follower pair M10,11 and the common-emitter pair M12,13,
From the noise and linearity analysis, it is evident that using shown in Fig. 7(b), are used to subtract the two second-
the subtraction of two second-order filters, to achieve fourth- order LC tank outputs. This gives a fourth-order BPF response
order filtering, results in the same noise figure (NF) as that of by matching the gains from the two LC tank outputs using
the individual second-order filters (see the circuit level noise resistance R E . The optimized value for R E is 22
. Resistance
AMIN et al.: INTEGRATED SYNTHETIC FOURTH-ORDER Q -ENHANCED BPF 775
Fig. 8. Varactor control scheme. (a) Typical single continuous mode control with the single large non-linear varactor. (b) Four independent controls employing
three switched and one continuous mode control to improve linearity by reducing the non-linear capacitance. (c) Simulation results of frequency coverage
of the continuous single varactor control and split four-varactor control schemes. (d) LC tank linearity with continuous single varactor control and split
four-varactor control—three switched varactors control plus one continuous varactor control.
R0 = 26
and the emitter-follower pair together serve as inductor of L = 325 pH, tunes the tank center frequency
matching to the 50-
load for measurement purposes. from 4 to 8 GHz including all parasitic effects. Varactor
capacitance is a function of control voltage VGS across its
A. Q-Tuning of Second-Order BPF terminals. The C–V relation depicted in Fig. 8(a) for a single
large varactor is linear or quasi-linear toward the minimum
The variable resistor RC in Fig. 7(a) is composed of linear
and maximum capacitance regions A and strongly nonlinear
350
resistors and NMOS Ms1,s2 in triode region, as shown
in the transition region B versus control voltage. In [35],
in Fig. 7(b). By controlling the resistance RC , Q of the second-
to improve the varactor linearity, a dual-varactor inverse (DVI)
order BPF can be increased up to ∼100 and reduced to as low
control scheme is adopted where two varactors are biased with
as ∼4. The parallel combination of Ms1,s2 and RC , in series
different operating points so that the nonlinearity from each
with ∼20-
output impedance of M1,2 , forms the LC tank
varactor can be partially canceled out by each other. This
load. This load varies from 27 to 370
and controls √ each method requires manual optimization or lookup-table-based
resonator’s intrinsic filter Q(= R P /Z o where Z o = L/C)
bias control to find optimum bias point of each varactor.
from 3 to 42 at 4 GHz. Transistors M1,2 and M9 set dc
More systematic linearity improvement can be achieved by
bias across resistance RC to be equal (VCC –VBE,on ). This
a switched varactor control with a number of varactor array
forces Ms1,s2 into deep triode region to operate as a linear
with a smaller size. This is illustrated in Fig. 8(b). By dividing
controlled resistor. The parasitic LC tank loss caused by a
the single varactor into four equally sized smaller varactors
finite Q of inductor and varactors can be compensated by the
where the three varactors can be switched in between the
negative resistance cell R N . R N is formed by the degenerated
quasi-linear regions A, the nonlinear varactance range B can
Darlington pairs M5−8 , R X , and RY for improved linearity
be shrunk by four times. Furthermore, thanks to the reduced
and minimal center frequency drift during Q-tuning [35], [36].
nonlinear capacitance range, the varactance transition slope is
Transistors M5,6 and R X = 200
give fixed linearized
also scaled down by the same factor, resulting in much milder
negative resistance, where R X also traps the transistor noise.
nonlinear behavior during the varactance transition region.
Additional variable negative resistance is achieved from the
This can be verified from SPECTRE periodic steady-state
transconductor pair M7,8 and variable resistance R X . The
(PSS) simulation. Fig. 8(c) shows the simulation results
total R N can be approximated as −2(R X ||RY ) differentially.
of the coverage of LC tanks resonance frequency using
the single varactor and four individual varactors control,
B. Switched Varactor Frequency Tuning respectively. The corresponding nonlinearity of the LC tanks
Fig. 8 shows that the LC tank consists of four differential in terms of input 1-dB gain compression point (IP−1 dB )
paired NMOS varactors and inductor L, utilized as a frequency is disclosed in Fig. 8(d). In the simulation, L = 325 pH
tuning element in this paper. The total varactor tuning capac- and overall varactance range are Cvar = 1.12–4.2 pF
itance (Cvar ) range is 1.1–4.2 pF which, together with an (4 · W/L = 1008/0.37, GF8HP model name: ncap). The
776 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 54, NO. 3, MARCH 2019
approach of switching ON and OFF three varactors, while the [see Fig. 7(b)]. Thus, the effective noise from the negative
fourth varies continuously guarantees coverage of the entire resistance unit is comprised of the thermal noise of RY and the
frequency range with no discontinuity [Fig. 8(c)]. NMOS channel noise. The mathematical details on the noise
In Fig. 8(d), it is clearly observed that during the varactor analysis are provided in Appendix B and the final outcomes
transition region, from roughly 4.25 GHz to around 7.5 GHz, of noise factor (F) of filters A and B are expressed as
the linearity degrades severely in the continuous single varac- (20) and (21), respectively
tor control case, albeit there is a sweet spot around 5.7 GHz
which is near the midpoint of the varactor transition. In the F|Filter
A
⎛ ⎞
switched varactor control scheme, case ④ where {Vc1 , Vc2 ,
Vc3 } = {VL , VL , VL } exhibits the steepest frequency transi- 2 ⎜ 1 R0 ⎟
RG ⎜ 1 ⎟
tion. However, the minima of IP−1 dB is 6–8 dB better than ≈ 2+4 ⎜ + γ (gmn + gdsn ) + + 2 2 ⎟
Rs ⎝ RG RY ⎠
those of the single varactor case, resulting in close to or better Q 1Z o1
Gm Negative Resistance Subtractor
than 0-dBm IP−1 dB for most of frequency range, except for
the high-frequency edge. Both varactor control schemes are (20)
essentially identical at the edges of the frequency range where F|Filter
B
IP−1 dB performances are about the same for both control ⎛ ⎞
schemes in Fig. 8(d). The switched varactor array approach 2 ⎜ 1 RE ⎟
RG ⎜ 1 ⎟
with a larger number of small varactors can improve the ≈ 2+4 ⎜ + γ (gmn + gdsn ) + + 2 2 ⎟.
Rs ⎝ RG RY ⎠
linearity further, and overall Q-enhanced LC-tank nonlinearity Q 2Z o2
would then be limited by the negative resistance R N nonlin- Gm Negative Resistance Subtractor
earity. (21)
In (20) and (21), γ , gmn , and gdsn are the NMOS channel
C. Circuit Noise Analysis thermal noise coefficient, transconductance, and channel con-
The noise analysis at the circuit level is performed for ductance, respectively. Z o1 and Z o2 , respectively, express the
the two paths from the filter A and filter .
B Fig. 9(a) filters
A and B LC tank characteristic impedance. Q 1 and Q 2
shows the major noise sources in the filter .
A The transistor are the quality factor of the LC tank of filters A and B
noises from M1 , M3 , and M10 cause a minor effect on including negative resistance.
the overall noise performance because of their low emitter Fig. 9(b) shows the plots of (20) and (21) with typical
impedance compared with loading impedances. Major noise design values at the center frequencies of f 01 and f 02
contributors are resistor thermal noises from the input and along with circuit simulation results for a fourth-order
output matching resistors RI and Ro , the transconductor gain fractional BW of 2%–25%. For this fractional BW range,
control resistor RG , the Q-tuning resistor RC , the LC tank Qs (Q 1 and Q 2 ) vary from 62 to 7. The individual filter’s NF
parasitic resistor RP , and its compensation resistor RN . In a at f 01 and f 02 matches well with the theory. The simulated
negative resistance cell, the variable resistor RY is realized noise figure of the resultant fourth-order filter at the center
with a parallel of linear resistor RY and NMOS transistor frequency f c = ( f01 + f 02 )/2 is also plotted and shows that
AMIN et al.: INTEGRATED SYNTHETIC FOURTH-ORDER Q -ENHANCED BPF 777
D. Mismatch Consideration Fig. 11. Chip photograph (1.05 × 0.95 mm2 including pads).
Ideally, the two BPF BWs should be equal to synthesize a
fourth-order shaping. However, the technique is very robust
against mismatches between the BWs. Fig. 10(a) and (b) capacitance variation of 1.6 pF from 1.9 to 3.5 pF, which is
shows the simulation results with BW mismatches up to 30%. not realistic in practical designs. The less than ±2% error
The mismatch has a negligible impact on the ripple magnitude. corresponds to <100 fF of capacitance variation, which is
As far as the OOB filtering is concerned, the roll-off does not a more realistic case for the parasitics effect. For this case,
change significantly for a 15% mismatch—namely, as can be the maximum ripple variation is 0.5 dB and the fourth-order
seen in Fig. 10(a) and (b), the BW mismatch up to 15% causes 3-dB BW variation is around 110 MHz, ∼11% of 1 GHz.
slight better roll-off on one side of center frequency with To control the tuning frequencies with precision, a closed-loop
slight degradation on the opposite side of the center frequency. control scheme could be adopted for frequency tuning [39].
From (2), a large mismatch introduces non-negligible zero In the linearity simulations, the IIP3 of the input buffer
which creates a notch either on the lower or higher side of the with >350-
loading is greater than 30 dBm. This, however,
BPF stopband depending on the two second-order resonators decreases to 16–19 dBm depending on the gain setting after
relative BWs. When the BW of filter A (ω01 /Q 1 ) is smaller adding the gain control transconductor. Due to the nonlin-
than the BW of filter (ω02 /Q 2 ) by 30%, a notch appears
B earity of frequency and Q-tuning elements, the IIP3 finally
on the lower side of the bandpass as can be seen in Fig. 10(a). dwindles to 8–11 dBm range after the varactor and negative
For the mismatch case when ω02 /Q 2 is smaller than ω01 /Q 1 , transconductor. Note that the filter is to be integrated on-chip
the notch is present at the higher side of the BPF. The and loaded by a large impedance, usually a transistor gate
advantages and disadvantages of BW mismatch are subjective impedance of the following building block. Under such large
to the requirement. By controlling the BWs, an intentional impedance loading, the IIP3 degradation by the subtractor is
notch can be introduced for blocker rejection for a specific negligible in the simulations.
interference scenario.
The effect of the error in second-order BPFs center fre-
quency (ω01 ) on the fourth-order BW and ripple magnitude IV. M EASUREMENT R ESULTS
is plotted in Fig. 10(c) for 1-GHz BW at 5.7-GHz center
frequency with less than 0.5-dB ripple. When ω01 increases Fig. 11 shows the fabricated chip micrograph of the
toward ω02 (positive %-error), the ripple reduces as well fourth-order BPF in the GF8XP process. The area of core
as the 3-dB BW. On the other hand, when ω01 decreases design excluding pads is 0.53 × 0.7 mm2 . The input variable
away from ω02 (negative %-error), the ripple and the 3-dB resistance Ri in Fig. 7(b) is controlled to measure matched
BW increases. The ±10% error in f 01 corresponds to the as well as large actual load cases.
778 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 54, NO. 3, MARCH 2019
Fig. 12. Measured (a) filter response at 6 GHz with fractional bandwidth tuning from 2% to 25% and (b) group delays at 6 GHz.
Fig. 13. Measured (a) NF, in-band P−1 dB , and dynamic range and (b) 3-dB group delay variation and phase variation.
A. BPF Characteristics at 6 GHz phase variation within 3-dB BW are plotted in Fig. 13(b) ver-
The measured filter performance for the BW tuning at the sus fractional BW. The phase variation within the 3-dB BW is
center frequency of 6 GHz is plotted in Fig. 12 for the input θ (ω)3 dB = ω3 dB × τ (ω) (22)
matched case, i.e., Ri = 50
. Fig. 12(a) shows the achieved
BW tuning from 2% to 25% with a passband ripple of less than where τ (ω) is the group delay in seconds and ω3 dB is the BW
0.5 dB. The corresponding measured group delay is shown of the filter. The group delay variation varies from 0.41 ns
in Fig. 12(b). The worst case 3-dB group delay is 410 ps. at 2% fractional BW to 0.23 ns at 25% fractional BW. The
Fig. 13(a) shows the noise measurements, carried out for both corresponding passband phase variations are 17.6° and 124°
input matched and high-impedance conditions. The measured for 2% and 25% fractional BW, respectively.
NF is 7.6 and 18 dB for the fractional BWs of 25% and 2%, Fig. 14(a) and (b) shows the measured OOB linearity tests
respectively, when Ri = 1 k
. For smaller fractional BW, at a center frequency of 6 GHz for fractional BWs of 2%,
noise power from the negative resistance cell increases with 5%, 10%, and 25%, corresponding to 120-, 300-, 600-, and
the increase in filter Q. The degradation in NF is consistent 1500-MHz BWs, respectively. The tests were performed
with the sensitivity-selectivity tradeoff in Q-enhanced for both upper and lower offset frequencies from the center
LC filters. For the input matched case, i.e., Ri = 50
, frequency of 6 GHz. The blocker P−1 dB test is shown
the NF is increased by 4–5 dB over the high-impedance case in Fig. 14(a), where a large blocker is injected at offset
due to the resistive matching halving the input signal power. frequency and a small signal is at the center frequency of
As discussed in Section III, the filter employs a variable 6 GHz. The blocker power is increased until the in-band signal
gain transconductor, and by increasing the gain, the filter NF compresses by 1 dB and is plotted for upper and lower sides.
can be reduced substantially. This comes at the expense of Similarly, the OOB IIP3 is performed for upper and lower side
linearity degradation. For the matched input case, the in-band offset frequencies from 6 GHz and is plotted in Fig. 14(b).
IP−1 dB is −1 ∼ 4 dBm. The resulting normalized DR, shown
in Fig. 13(a), is 151–166 dB for the factional BW range B. Frequency Tuning
of 2%–25%. The normalized DR = 174+IP−1 dB −NF [dB] is The measured center frequency tuning from 4 to 8 GHz
the ratio of in-band 1-dB compression point and input-referred is shown in Fig. 15(a) for a constant fractional BW (equiv-
noise floor in a 1-Hz BW [29]. The group delay variation and alent Q = 20). The passband ripple is less than 0.2 dB.
AMIN et al.: INTEGRATED SYNTHETIC FOURTH-ORDER Q -ENHANCED BPF 779
Fig. 14. (a) Blocker P−1 dB measurements for 2%, 5%, 10%, and 25%. (b) OOB IP3 measurements for 2%, 5%, 10%, and 25%.
Fig. 15. Measured (a) frequency tuning with constant fractional BW (Q = 20) and (b) frequency tuning with absolute BW of 200 MHz.
Fig. 16. Measured (a) NF, in-band P−1 dB , and DR for constant fractional BW and (b) NF, in-band P−1 dB , and DR for absolute BW.
The filter center frequency tuning is continuous, but for have an absolute BW response, as shown in Fig. 15(b).
reporting purposes, is plotted here for discrete frequency steps. In this case, the BW at each center frequency is 200 MHz
The filter has an excellent out-band-rejection on the lower and the passband ripple is less than 0.2 dB. The filter has
side away from the center frequency and is more than 60 dB more than 60 and 45 dB of rejection toward the lower and
at 1 GHz. On the higher frequency side, the OOB rejection upper side frequencies, respectively. The measured NF and
is more than 42 dB and saturates due to the high-frequency in-band IP−1 dB for the constant fractional BW are plotted in
zero caused by the mismatch in BWs of each resonator and Fig. 16(a) for the input matched case of Ri = 50
. The
dc blocking capacitors. The filter can also be controlled to gain is kept constant at 5 dB and the S21 plot is normalized
780 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 54, NO. 3, MARCH 2019
Fig. 17. Measured (a) gain control at 6 GHz and (b) NF, in-band P−1 dB , and DR versus gain.
TABLE I
P ERFORMANCE C OMPARISON W ITH THE S TATE - OF - THE -A RT BPF S
to 0 dB. The measured DR is 153.2 ± 1 dB · Hz for the gain tuning centered at 6 GHz for both input-matched and
entire frequency range of 4–8 GHz, as shown in Fig. 16(a), high-impedance cases. The corresponding measured NF and
and remains constant since Q of the BPF remains constant. IP−1 dB are plotted in Fig. 17(b). The DR range varies from
The measured NF and in-band IP−1 dB for the absolute BW 153.2 to 158.6 dB. The measured filter performance and
of 200 MHz are plotted in Fig. 16(b). The DR varies from its comparison to the state-of-the-art integrated circuit filters
154.6 to 147.3 dB and 4 to 8 GHz, respectively. The DR are summarized in Table I. The proposed fourth-order filter
degrades at high frequencies since more Q is required to achieves one of the widest tunable BW ranges (from 120 MHz
maintain constant BW of 200 MHz, in turn requiring more to 1.5 GHz at 6 GHz), better DR, and higher OOB rejection,
negative gm and hence increased NF. making this design favorable for operation at microwave
frequencies. The power consumption of the filter is on par
with, or better than, the N-path approach at C-band and
C. Gain Tuning beyond.
The filter gain can be tuned based on the NF and linearity Fig. 18 summarizes the power consumption contribu-
tradeoff due to the current-mode driving using the variable tions from different blocks for a power supply of 3.5 V.
transconductor in Fig. 7(b). Fig. 17(a) shows the measured Fig. 19 shows the power consumption of the published N-path
AMIN et al.: INTEGRATED SYNTHETIC FOURTH-ORDER Q -ENHANCED BPF 781
# 1 R0 Ri
1
+ 1
−1 + + 2 + 2
RTank . (A15)
ω Mr Mr2 RY RTank 4RG2
= (A7)
ωR B 2 From (A15), it is straightforward to calculate the noise factor
where Mr (ωc ) is simply notated as Mr . of filter
A which is found to be
When ripple magnitude Mr is in dB, (A7) can be expressed
F|Filter
A
as
ωR B 2
Vn,overall
ω = (10 Mr /20 + 10 Mr /10 − 1). (A8) = 1+
2 4kT Rs A2v
⎛
The 3-dB BW of fourth-order BPF can be written as
# ⎜ 1
√ 2
RTank ⎜ 1 1 1
ω3 dB = 2ω + ω R B 2−1 = 2+ ⎜ + + + γ (gmn + gdsn ) +
# Rs A2v ⎝ RG RP RC RY
√
= ω R B (10 Mr /20 + 10 Mr /10 − 1 + 2 − 1). (A9) Gm Q-enhanced LC Tank
⎞
From (A9), Q of individual resonator, ω and resonance ⎟
frequencies ωo1 and ωo2 in Fig. 3(a) can be expressed in terms R0 ⎟
+ ⎟
of 3-dB BW (ω3 dB ) and passband ripple magnitude (Mr in 2
RTank ⎠
dB-scale) as (3)–(5), respectively, in the main text. Subtractor
⎛ ⎞
A PPENDIX B 4R 2G ⎜ 1 R0 ⎟
⎜ 1 ⎟
≈ 2+ ⎜ + γ (gmn + gdsn ) + + 2 2 ⎟
In Fig. 9(a), when all the noise powers are referred to the Rs ⎝ RG RY ⎠
Q 1Z o1
LC tank output, the net noise power density from the LC tank Gm Negative Resistance Subtractor
2
(Vn,Tank ) including negative resistance cell is expressed as (A16)
⎛ ⎞
where, RTank = Q 1 Z o1. Q 1 is the quality factor of the LC tank
⎜ 1 1 ⎟ of filter
A including negative resistance. Z o1 is the√filter A
⎜ 1 ⎟ 2
2
Vn,Tank = 4kT ⎜ + + γ (gmn + gdsn ) + ⎟R LC tank characteristic impedance and expressed as L/Cvar .
⎝ RP RC RY ⎠ Tank
Following the same procedure, the noise factor of filter B
γEq G m1,2 in (6)
found to be
(A10)
F|Filter
B
where, γ , gmn , and gdsn are the NMOS channel thermal ⎛ ⎞
noise coefficient, transconductance, and channel conductance, ⎜ 1
respectively. The negative resistance noise of γEq G m1,2 in (6) 4R 2G ⎜ 1 RE ⎟ ⎟
≈ 2+ ⎜ + γ (gmn + gdsn ) + + 2 2 ⎟.
is γ (gmn + gdsn ) + 1/RY in actual circuit implementation. Rs ⎝ RG RY Q Z ⎠
2 o2
Following are the noise powers from other major dominant Gm Negative Resistance Subtractor
sources. (A17)
2
Variable gain transconductor noise (Vn,Gm ):
Q 2 is the quality factor of the LC tank of filter
B including
4kT 2 negative resistance. Z o2 is the filter
B LC tank characteristic
2
Vn,Gm ≈ R . (A11)
RG Tank impedance.
AMIN et al.: INTEGRATED SYNTHETIC FOURTH-ORDER Q -ENHANCED BPF 783
Farooq Amin (S’14–M’18) received the B.S. degree Kwang-Jin Koh (S’06–M’09) received the Ph.D.
in computer engineering from COMSATS University degree in electrical and computer engineering
Islamabad, Islamabad, Pakistan, in 2005, the M.S. from the University of California at San Diego,
degree in electrical engineering with specializa- San Diego, CA, USA, in 2008.
tion in system-on-chip from Linköping University, From 2000 to 2004, he was a Research Staff with
Linköping, in 2009, and the Ph.D. degree in electri- the Electronics and Telecommunications Research
cal engineering from Virginia Polytechnic Institute Institute, Daejeon, South Korea. From 2008 to 2010,
and State University (Virginia Tech), Blacksburg, he was a Senior Engineer with Intel Corporation,
VA, USA, in 2017. Hillsboro, OR, USA. From 2010 to 2011, he was a
From 2011 to 2013, he was a Design Engineer with Senior Staff Scientist with Broadcom Corporation,
the CPU Design Group, Intel Corporation, Hillsboro, Irvine, CA, USA. From 2012 to 2018, he was an
OR, USA, where he was working on the Intel Hasewell and Broadwell Assistant Professor with the Electrical and Computer Engineering Depart-
family of microprocessors design. In 2016, he joined the Qorvo Advanced ment, Virginia Polytechnic Institute and State University (Virginia Tech),
Technology Group, Greensboro, NC, USA, as an intern, where he was working Blacksburg, VA, USA. His current research interests include analog and
on the next-generation SOI RF switch EM modeling. He is currently an RFIC mixed-mode circuits, antennas, and applied electromagnetics from RF to
Design Engineer with Northrop Grumman Mission Systems, Baltimore, MD, millimeter-wave and terahertz range.
USA, where he is working on microwave and mm-wave T/R modules. His Dr. Koh has served as a Technical Program Committee Member for the
research interests include RF and mm-wave circuits, tunable filters, analog IEEE Bipolar/BiCMOS Circuits and Technology Meeting, the IEEE Custom
X-ray read-out circuits and systems, and low-power high-speed custom Integrated Circuits Conference, the IEEE/ MTT-S International Microwave
datapath design. Symposium (IMS), and the IEEE BiCMOS and Compound Semiconductor
Dr. Amin was a recipient of the IEEE MTT-S International Microwave Integrated Circuits and Technology Symposium. He and his doctoral advisees
Symposium Best Student Paper Award (First Place) in 2016, the were a recipient of the First Place Best Paper Award in the Advanced Practice
2014 and 2016 MTT Student Travel Grants, and the NASA-Virginia Space Paper Competition, the Honorable Mention Award in the Student Paper
Grant Consortium (VSGC) Graduate Fellowship from 2015 to 2017. Competition in the 2017 IEEE/MTT-S IMS, the First Place Best Student Paper
Award in the 2016 IEEE/MTT-S IMS, and the Second Place Best Student
Paper Award in the 2015 IEEE/MTT-S IMS. He was a recipient of the Best
Paper Award of the IEEE Solid-State Circuits and the IEEE Electron Device
Societies, Seoul Chapter in 2002, the Science, Technology, Engineering,
Sanjay Raman (S’84–M’98–SM’06–F’12) was and Mathematics Research Fellowships from the National Aeronautics and
born in Nottingham, U.K., in 1966. He received the Space Administration and the Commonwealth of Virginia from 2015 to 2017,
B.S. degree (Hons.) in electrical engineering from the Outstanding Assistant Professor Award from the College of Engineering
the Georgia Institute of Technology, Atlanta, GA, in 2014, and the Junior Faculty Research Award from the Institute for Critical
USA, in 1987, and the M.S. and Ph.D. degrees Technology and Applied Science in 2012, Virginia Polytechnic Institute and
in electrical engineering from The University of State University. He was a finalist of the R.W.P. King Award of the IEEE
Michigan at Ann Arbor, Ann Arbor, MI, USA, Antenna and Propagation Society in 2015. He was also one of the recipients
in 1993 and 1998, respectively. of the Team of the Year Award in 2010 from the Teledyne Technology Inc.,
From 1987 to 1992, he was a Nuclear Trained (formerly, Rockwell Scientific Corporation) in recognition of his microwave
Submarine Officer with the U.S. Navy. In 1998, and millimeter-wave silicon phased arrays which were reported to the U.S.
he joined the faculty of the Bradley Department Department of Defense by the Defense Advanced Research Projects Agency
of Electrical and Computer Engineering, Virginia Polytechnic Institute as one of the agency’s major achievements from 2007 to 2008.
and State University (Virginia Tech), Blacksburg, VA, USA, where he is
currently a Professor. From 2007 to 2013, he served as a Program Manager
with the Microsystems Technology Office, Defense Advanced Research
Projects Agency (DARPA), Arlington, VA, USA, where he was responsible
for major Research and Development programs in the areas of high-speed
electronics, adaptive RF/mixed-signal integrated circuits, RF MEMS, and
3-D/heterogeneous integration technologies. He is also currently serving
as an Associate Vice President for the Virginia Tech National Capital
Region, where he is responsible for planning and executing region-wide
initiatives to enhance the university’s research, education, and outreach
missions and the President and CEO for the Virginia Tech Applied Research
Corporation. He is also a founding member of the VT Multifunctional
Integrated Circuits and Systems (MICS) Research Group. His research
interests include RF/microwave/millimeter-wave integrated circuits and
antennas, high-speed/mixed-signal ICs, interconnects and packaging,
RF microelectromechanical/nanoelectromechanical (MEMS/NEMS) devices,
and integrated wireless communications and sensor microsystems.
Dr. Raman was a recipient of the 2007 Virginia Tech College of
Engineering Faculty Fellow, the 2000 Presidential Early Career Award
for Scientists and Engineers (PECASE), the 1999 NSF CAREER Award,
the Virginia Tech College of Engineering Outstanding New Assistant
Professor Award in 2000, the Armed Forces Communications and Electronics
Association (AFCEA) Postgraduate Fellowship from 1996 to 1997, and the
Secretary of Defense Medal for Exceptional Public Service for his service
at DARPA in 2013. He has served as an Associate Editor for the IEEE
T RANSACTIONS ON M ICROWAVE T HEORY AND T ECHNIQUES and on the
Technical Program Committee of the IEEE Radio Frequency Integrated
Circuits Symposium. He served as the Technical Program Co-Chair for the
2014 International Microwave Symposium. He is an Elected Member of the
IEEE Microwave Theory and Techniques Society (MTT-S) Administrative
Committee.