CT ICfilters
CT ICfilters
Integrated Filters
Types of Filters
• A continuous time filter is a filter whose variables are continuous both in time and in amplitude.
• A discrete-time (analog sampled-data) filter is a filter whose variables are continuous in amplitude but not
in time.
• A digital filter is a filter whose variables are discrete both in time and in amplitude.
Example:
8
7
Continuous Sampled and held
6
time signal analog value
Amplitude
5
4
3
2
1
0 1 2 3 4 5 6 7 8
Sample times
__t
T Fig.11.1-02
Digital Filters
Current technology constrains digital filtering to 5-10MHz or less.
where
1
H. Khorramabadi and P.R. Gray, “HIgh Frequency CMOS Continuous-Time Filters”, IEEE J. Solid-State Circuits, pp. 939-948, Dec. 1984.
2
G. Groenewold, “The Design of High Dynamic Range Continuous-Time Integratable Bandpass Filters,” IEEE Trans. Circuits and Systems, pp. 838-
852, Aug. 1991.
†
W.B. Kuhn, “Design of Integrated, Low Power, Radio Receivers in BiCMOS Technologies,” Ph.D. Dissertation, Electrical Engineering Dept., Virginia
Polytechnic Institute and State University, Blacksburg, VA, 1995.
All designs start with a normalized, low pass filter with a passband of 1 radian/second and an impedance of
1Ω that will satisfy the filter specification.
1.) Cascade approach - starts with the normalized, low pass filter root locations.
2.) Ladder approach - starts with the normalized, low pass, RLC ladder realizations.
-bN
-b3
-b2
-b1
Vin Σ T1 T2 T3 TN
a0 a1 a2 a3 aN
Σ Vout Fig.11.2-01
†
A.S. Sedra and P.O Bracket, 1978, Filter Theory and Design: Active and Passive, London: Pitman.
Vin Σ T T T T Vout
Fig.11.2-02
Note that the design of switched capacitors is done in such a manner that denormalization is not necessary.
-
+
Active Filters
Passive Filters
Fig.11.1-03
Passive Filters
Categories:
Discrete ceramic (piezoelectric)
Crystal
Acoustic wave
Surface (SAW)
Bulk (BAW)
LC
General Characteristics:
• Fractional bandwidths are small (0.1% to 3%)
• Shape factors are moderate (16 to 20dB of attenuation at 2 to 3 times the nominal bandwidth)
• Insertion loss is moderate (1.5 to 6dB)
• Cost is low ($0.3 to $3) when purchased in large quantities
• No power dissipation
• Low noise figure
Counter-clockwise Clockwise
quadrature sequence quadrature sequence
R R
V 0° V 0°
C C
R V' 0°+45° R V' 0°+45°
V 90° + V 270° +
C V' 90°-45° C V' 270°-45°
= 2V' +45° =0
R R
V 180° V 180°
C C
R R
V 270° V 90°
C C
-70
-80
-90
-100
Four stages of polyphase filters in cascade with different center frequencies.
Fig10-PF4
• Wideband image rejection can be obtained with staggering several polyphase stages
• Wider polyphase filter band ⇒ Large number of stages ⇒ More loss
• Measurements on poly resistors of a 1µm CMOS process gave σ < 0.1% for a resistor area of 2800µm2
• Approximate matching of capacitors is σ ≈ 0.1% for a capacitor area of 220µm2
• Last polyphase stage can provide up to 3dB • Outputs load each other, therefore the
of gain with small loading capacitors maximum gain is 0 dB
• Dummy capacitors try to match the output • This is an order of magnitude less
loading on all branches sensitive to load mismatches if
• Because of the matching issue, this is useful CLoad<<CPoly and almost zero
in high image rejection systems only if sensitivity to load mismatch if
CLoad<<CPoly CLoad>CPoly
Zin = R||(1/sC) R
ZLoad
at f = 1/(2πRC) C
R
ZLoad
C
R
ZLoad
C
R
ZLoad
C
Fig. 10-PF6
• If several polyphase filters are in cascade, the noise of the previous stage is attenuated by the following.
Therefore, the last stage noise becomes dominant.
• There is a tradeoff between the loading of the polyphase filter on the preceding stage (which sets the loss)
and the noise of the polyphase filter which establishes the resistor.
8.) Polyphase Voltage Gain and Loss:
RPoly RPoly
Vout
Vout
Vin 0° CPoly ZLoad CPoly
Vin 90°
Fig. 10-PF7
Therefore,
ZLoad
|Vout| = 2 Z |Vin|
Load+ZPoly
a.) Output open circuit: ZLoad>>ZPoly ⇒ 3dB gain
b.) ZLoad = ZPoly ⇒ 3dB loss
Gradient Cancellation
C
R
Comments:
C
1.) Common centroid layout is used to cancel the process gradients R
Gradient Cancellation
R
C
R
C
R
C
R
C
Dummy
Elements R
Fig10PF-8
A B
C D
Fig.10-PF9
To
Ground Fig.10-PF10
SECTION 3 - INTEGRATORS
The Role of the Integrator in Active Filters
In most active filters, the summing integrator is the key building block (the primitive).
v1 k1
v2 k2 Σ vo
vn kn Fig.11.4-01
k1 k2 kn
Vo(s) = ± s V1(s) ± s V2(s) ··· ± s Vn(s)
Classical realization:
Damping Resistor
C
Vout(jω) -1 R
H(jω) = = Vin
Vin(jω) jωRC Vout
Integrator Nonidealities
dB Phase Shift
180°
Ao
R C 135°
Vin Vout
- GB log10ω 90°
0dB
+ ωa ωi 45°
ωi GB GB 10GB log10ω
0°
ωa ωa 10ωa
10 10 Fig10-10
Ao GB
A(s) = =
s s + ωa
1+
ωa
The integrator behavior is degraded by:
• Finite Ao in the low frequency region
• Finite GB in the high frequency region
Integrators - Continued
Biquad uses ±summing integrator with damping for second-order stages.
k1
k2 k4
Vin Σ Σ Vout
k3
+ - + -
Σ Σ
kS kL
Vin
+ Σ - + Σ -
Vout
What is a GM?
A transconductance amplifier with a
linear relationship between Iout and Vin. It must be tunable.
Iout
Iout
Imax
Imax
GM
1 Vin
Vin
-Imax
-Imax
Advantages: Disadvantages:
High frequency GM must be linear and tunable
Simple Sensitive to parasitic capacitances at the output nodes
Frequency Response:
Phase
dB Shift
180°
90°
0dB ω
1 ωI
RoC 0° ω
1 ωI
RoC
Fig.11.4-03
Solution:
Make Ro large by using cascode configuration
+ +
Gm(0)Vin
C Vout Vout -Gm(0)
Vin sτa+1 =
Vin sC(sτa,+1)
- -
OTA Fig.11.4-04
Frequency Response:
dB
Phase
Shift
90°
0dB ω
ωI 1
τa 0° ω
ωI 1
τa
Fig.11.4-05
Solution:
Use higher fT MOSFETs (smaller channel lengths)
Frequency Response:
Phase
dB Shift
180°
Gm 90°
Cz
0dB ω
ωI
0° ω
ωI 1
τa
Fig.11.4-07
Solution:
Neutralization techniques
Fig.11.4-08
OTA-C Integrators
VDD VDD VDD VDD
VDD VDD VB1
VB1 M3 M4
VB2
Voutn C Voutp
M3 Voutn Voutp
Vinp Vinn VB3
M1 M2
Vinp VC Vinn
M5 M1 M2
VC
VB4
Simple OTA:
Suffers from poor linearity, signal-to-noise limitations, and low output resistance
Folded-Cascode OTA (M3 in active region):
gm1 1 W3
Gm = 1+ g R ≈ R = Kn’ L (VC-VS3-VTn)
m1 M3 M3 3
†
F. Krummenacher, and G.V. Ruymbeke, “Integrated Selectrivity for Narrow-Band FM IF Systems,” IEEE J. os Solid-State Circuits, vol. SC-25, no. 3,
pp. 757-760, June 1990.
ID
I+Iout I-Iout
I+Iout I-Iout
M1
M2 QD
+ Q1 Q2
M3
Vin
M4 + RD
-
Vin M1 M2
I I -
I
VSS
VSS
(a.) (b.)
(a.) M3-M4 work in a saturation-active mode for positive Vin and in an active-saturation mode for
negative Vin. Can result in a linear operation. I varies GM. (b.) M1 and M2 are in active region. Current
is proportional to VDS1 (VDS2). ID varies GM.
- V + - V +
out out
VB3 VB3 VB3 VB3
VC Vinp VC Vinn
Vinp Vinn + +
M1 M2 M1 M2
- -
VB4 M3 VB4 M3
VB4 VB4
M2 M1
vout
M4 C M3
vin V2 V1
M6 M5
Fig. 11.4-9
GM-C-OTA Integrator
Integrator:
2C
Iout
+ - + +
Vin G OTA Vout
- + M- -
Iout
2C
Advantages:
Avoids influence of parasitic capacitances.
High frequency (20MHz, 6th-order filter for disk-drive read channels)
Tuning can be done by current multiplication of Iout
Disadvantages:
More noise - two active elements (GM and OTA)
i1 R
v1 - + + - + +
Vout VC2 Vout
v2 + - - + - -
i2
R
2C iD3 2C
iD4
v2
i2
VC1
Differential input resistance is (see Sec. 4.2):
v1-v2 v1-v2 1
Rin = 2R = i -i = = , v1,v2 ≤ min{(VC1-VT),(VC2-VT)}
1 2 β (VC1-VC2)(v1-v2) β (VC1-VC2)
Advantages:
Straight-forward implementation of RC active filters
Insensitive to parasitics
Disadvantages:
Requires input currents
MOSFET-C-OTA Filters
Integrator:
Replace the op amp of MOSFET-C-OP AMP filters with a high-gain OTA. At high frequencies, if the
GM is large, the circuit behaves as an op amp.
2C
i1 R
v1 - + +
OTA Vout
v2 + - -
i2
R
2C
K’ + - K’ K’ VxVy VxVy K’
Vout = sC (Vo - Vo ) = sC (+I4 - I3 - I1 + I2) = sC 2 + 2 = sC VxVy
K’
∴ Vout = sCVinVC
As long as the following conditions are satisfied, the integrator will be linear.
Vinp,Vinn ≤ min{(Vy+-VT),(Vy- -VT)}
K’/sC
∴ Vout = 1 2K’/sC VinVC
1 + A + A (VGS - VT)
K’/sC
∴ Vout = 1 2K’/sC K’/sC
VinVC
+ -
1 + A (VGS - VT) - 1 + A (∆VT + ∆VT )
1 + 1 +
A
Advantages:
Good linearity
Wide dynamic range
Disadvantages:
Op amp must be able to drive resistances
Requires large area
Analysis: Ib
Note that VBE1 + VBE2 = VBE3 + VBE4 iin iout
⇒ i1i2 = i3i4 ⇒ iinIb = i3iout Q2 Q3
dvBE4 Q1 Q4
Idamp
But, i3 = iC + Idamp = C dt + Idamp Ib C
and
dvBE4 d iout CVT diout CVT Logarithmic Exponential
C = CV Log domain integrator
dt T dt ln I = i = i s·iout compressor
out dt
expander
s out
Fig. 10-15
∴ iinIb = (sCVT + Idamp)iout
Ib
iout Ib CVT
⇒ i = sCV + I = Idamp (set Idamp = 0 for true integrator)
in T damp
s + CV
T
Comments:
• The corner frequency of the filter depends on temperature (tuning is required)
• Lack of pure logarithmic VI characteristics and finite β of the BJTs lead to linearity degradation
• Good for high frequency (100MHz)
Io Ip+ Ip- Io
Iout+ Q2 Q1 Q7 Q8
Iout-
V+ V-
Vout+ Vout-
Io Io
C C
Ip+ Iout
In+ In- In+ Vout+
Io Io V+
Q5 Q11 Iout
Vout-
Q4 Q3 Q9 Q10 V- In-
Ip-
C C
Q6 Q12
R R
Fig. 10-LD1
I I I I KI
ip if iout
in a
+
C va
1:1 - 1:1 1:K
in-ip- if
Thus,
iout = Kif = (ip - in)
Use the dc current sources (I and KI) to tune the transconductances.
Analysis gives,
gm
4gds 1 + sgdsC2i - 1 - sC1i
if = p n
sC2 sC1 g m2 gm
1 + 1 +
4gds gm
• Dominant pole moves from the origin to -4gds/C2
• Unity gain frequency is gm/C2
• The undesired capacitance, C1, inversely affects the frequency of the non-dominant pole (gm/C1)
causing an undesirable excess phase shift at the integrator unity gain frequency
• If C2 >> C1, and one uses cascode techniques to reduce the output conductances, Q factors exceeding
20 can be achieved at high frequencies
Ideally, VDD
gm
iop - ion = sC (ip - in) ip I I I ion
In reality,
iop
s C
1 -
z1
iop - ion = A (i - i )
s p n
1 + VDD
p1
where I I I
ion iop
gm - gds in
A= gds
C
gm - gds
z1 = 2C
gd
and
gds
p1 = C + 4C
gd
SECTION 4 - BIQUADS
The Biquad
A biquad has two poles and two zeros.
Poles are complex and always in the LHP.
The zeros may or may not be complex and may be in the LHP or the RHP.
Transfer function:
Vout(s) -(K2s2+ K1s + K0) (s-z1)(s-z2)
Ha(s) = V (s) = = K (s-p )(s-p )
in ωo 1 2
s2 + Q s+ ωo2
jω
ωo
σ
ωo
2Q
K1 Fig.11.5-02
K1 Fig.11.5-03
More Biquads
KHN:
-1 Vlp(s)
-KLP
Vin(s) Vhp(s) Vbp(s) Vlp(s) Vbp(s) Vout(s)
-1 -1
K1 Σ ωo
s
ωo
s
KBP Σ
Vhp(s)
KHP
1/Q
Fig.11.5-04
Tow-Thomas Biquad:
ωo
ωo/Q
Fig.11.5-05
OTA-C BIQUADS
Building Blocks for OTA-C Biquads
Vi1 + Vi1 +
Gm1 Gm1
- -
Vi1 + Vo Vi1 + Vo Vi2 + Vo Vi2 + Vo
Gm1 Gm1 Gm2 Gm2
Vi2 - Vi2 - - -
- -
C Gm2 Gm4 C
Vi3 + Vi3 +
+ Gm3 + Gm3
- -
Fig.11.5-06
-
Gm
+
C
Vout
+
Gmm1 -
- Gm/Q
C +
Vin -
Gm
+
Fig.11.5-13
Transfer function:
GM
Vout(s) C s
Vin(s) = GM GM2
s2 + QC s + C
+ -
C G
- M+
- + +
G C Vout
+ M- -
+ - + + -
Vin G GM/Q
- + M- - +
Transfer function:
GM
Vout(s) C s
Vin(s) = GM GM2
s2 + QC s + C
2CA 2CB
+
+ + + + + +
Gm1 Gm2 Gm3 Vout
- - - - - -
-
2CA 2CB
2Cz
+
+ +
Vin Gm4
- -
-
Fig.11.5-075
2Cz
Example 4-1
Use the biquad of the previous page to design the following transfer function where K2=4.49434x10-1, K0 =
4.580164x1012, ωo = 2.826590x106 and Q = 6.93765.
Vout(s) K2s2+ K0
Ha(s) = V (s) =
in ωo
s2 + Q s+ ωo2
Find the values of all transconductances and capacitors and let Ca =2pF, Cb = 35pF and Gm1=Gm2.
Solution
Equating Ha(s) to the previous transfer function gives
Cz Gm2Gm4 ωo Gm3 Gm1Gm2
K2 = C +C , K0 = C (C +C ) , Q = Cb+Cz , and ωο2 = C (C +C )
b z a b z a b z
K2Cb 0.44943·35pF
Thus, Cz = 1-K = 1-0.44943 = 28.2264pF
2
K0 Gm4 K0 4.580164x1012
Note that 2 = G which gives Gm4 = 2 Gm1 = 31.785µS = 18.221µS
ωo m1 ωo (2.826590x106)2
2CA 2CB +
+ + + + + +
Gm1 Gm2 Gm3 Vout
- - - - - -
2CA 2CB -
2CX
+
+ + + +
Vin Gm4 Gm5
- - - -
-
Fig.11.5-12
2CX
Transfer function:
Cx G Gm2Gm4
2 m5
s + s +
Vout Cx+CB Cx+CB
CA(Cx+CB)
Vin = Gm3 Gm1Gm2
s2 + C +C s + C (C +C )
x B A x B
- Va
Vin Gm1 + Vout
+ Gm2
C1 -
C2 -
Gm3
Fig.11.5-14
+
Analysis:
Σi=0 at Va ⇒ sC1Va-Gm1(Vin-Vout) = 0 → sC1Va = Gm1(Vin-Vout)
Σi=0 at Vout ⇒ sC2Vo2 + Gm3Vout - Gm2Va = 0
Combining equations gives,
Gm1(Vin-Vout)
sC2Vo2 + Gm3Vout - Gm2 sC
=0 → (s2C1C2 + sC1Gm3 +Gm1Gm2)Vout = Gm1Gm2Vin
1
Gm1Gm2
Vout Gm1Gm2 C1C2
∴ Vin = = Gm3 Gm1Gm2
s2C1C2 + sC1Gm3 +Gm1Gm2 2
s +s C + C C
2 1 2
An alternate approach:
+
Gm1(0)τa1- Gm2(0)τa2 Gm1 +
Gmeq = (Gm1(0)-Gm2(0)) 1 - s Gm1(0)-Gm2(0) - Gmeq
-
-
Gmeq = Gm1(0)-Gm2(0) if Gm1(0)τa1 = Gm2(0)τa2 Gm2 Fig.11.0-01
+
MOSFET-C BIQUADS
General Second-Order Biquad
G4
C1 G5
CA CB
G2
G1
-+ G3 -+
+ +
Vin + - G3 + - Vout
- -
G1 CA CB
G2
C1 G5
Fig.11.5-11 G4
Transfer function:
C1 G G1G3
2 2
CB s + CB s + CACB
Vout
Vin = - G G3G4
2 5
s + CBs + CACB
1
K.R. Laker and W.M.C. Sansen, Design of Analog Integrated Circuits and Systems, McGraw Hill, New York, 1994.
2
P.E. Allen and E. Sanchez-Sinencio, Switched Capacitor Circuits, Van Nostrand Reinhold, New York, 1984.
3
R. Gregorian and G.C. Temes, Analog MOS Integrated Circuits for Signal Processing, John Wiley & Sons, New York, 1987.
4
L.P. Huelsman and P.E. Allen, Introduction to the Theory and Design of Active Filters, McGraw Hill Book Company, New York, 1980.
Example 5-1 - Fifth-order, Low Pass, Continuous Time Filter using the Cascade Approach
Design a cascade, switched capacitor realization for a Chebyshev filter approximation to the filter
specifications of TPB = -1dB, TSB = -25dB, fPB = 100kHz and fSB = 150kHz. Give a schematic and
component value for the realization. Also simulate the realization and compare to an ideal realization.
Solution
First we see that Ωn = 1.5. Next, recall that when TPB = -1dB that this corresponds to ε = 0.5088. We
find that N = 5 satisfies the specifications (TSB = -29.9dB).
Find the roots for the Chebyshev approximation with ε =0.5088 for N = 5. Therefore we can express
the normalized lowpass transfer function as,
0.2895 0.9883 0.4293
TLPn(sn) = T1(sn)T2(sn)T3(sn) = s +0.2895 2 2 .
n sn+0.1789sn+0.9883sn+0.4684sn+0.4293
Next, we unnormalize these values using a value of Ωn = 105·2π and an arbitrary impedance scaling
of zo = 105. Thus, we get the following denormalizes values of
0.2895S 1F
Gm11 = Gm21 = 5 = 2.895µS and C11 = 5 = 15.9pF
10 10 ·2x105π
Stage 2 - Second-order, High-Q Stage
The next product of TLPn(sn) is -
Gm12 + Vout
Gm12Gm22 Vin
+ Gm22
C12
0.9883 C12C22 -
C22 -
T2(sn) = 2 = Gm32 Gm12Gm22 Gm32
sn + 0.1789sn + 0.9883 2 +
s +s C + C C Fig.11.6-02
22 12 22
Assume that Gm12n = Gm22n and C12n = C22n = 1F. Equating coefficients gives,Gm12n = Gm22n =
0.99413S and Gm32n =0.1789S,
Denormalizing with Ωn = 105·2π and zo = 105 gives Gm12 = Gm22 = 9.94133µS, Gm32 =1.789µS
Denormalizing with Ωn = 105·2π and zo = 105 gives Gm13 = Gm23 = 6.552µS, Gm33 =4.684µS
and C13 = C23 = 15.9pF.
Overall realization:
Vin - -
+ Vo1 Gm12 + Vo2 Gm13 + Vout
Gm11 + Gm22 + Gm23
- C12 - C13 -
C11 - C22 - C23 -
Gm21 Gm32 Gm33
+ + +
Fig.11.6-04
All capacitors are 15.9pF and Gm11 = Gm21 = 2.895µS, Gm12 = Gm22 = 9.94133µS, Gm32 =1.789µS, Gm13
= Gm23 = 6.552µS, and Gm33 = 4.684µS.
-10
-20
-30
dB
-40
-50
-60
10 100 1K 10K 100K 1M
Frequency (Hz) Fig.11.6-16
EXAMPLE 5-1 - OTA-C Cascade Filter R4 4 0 10000MEG
.OPTION LIMPTS=1000 G32 4 0 4 0 1.789U
VIN 1 0 DC 0 AC 1.0 G13 0 5 4 6 6.552U
R11 1 0 10000MEG C13 5 0 15.9P
G11 0 2 1 0 2.895U R5 5 0 10000MEG
C11 2 0 15.9P G23 0 6 5 0 6.552U
R2 2 0 10000MEG C23 6 0 15.9P
G21 2 0 2 0 2.895U R6 6 0 10000MEG
G12 0 3 2 4 9.94133U G33 6 0 6 0 4.684U
C12 3 0 15.9P .AC DEC 10 1 10MEG
R3 3 0 10000MEG .PRINT AC VDB(6) VP(6) VDB(4) VP(4) VDB(2) VP(2)
Specification Value
Lower Stopband 400 KHz
Lower Passband 439 KHz
Upper Passband 461 KHz
Upper Stopband 506 KHz
Passband Ripple <0.5dB
Stopband Attenuation >55dB
Tuning Resolution <1%
Power Consumption ≈2mA
Total Inband Noise <340µVrms
Cascaded-biquad BFP structure:
To compromise between the group delay and complexity, a Chebyshev approximation of 12th order is
used.
Vin Vout
Biquad 1 Biquad 2 Biquad 3 Biquad 4 Biquad 5 Biquad 6
Fig.11.6-06
2Ca 2Cb
+
+ + + + + +
Gm1 Gm2 Gm3 Vout
- - - - - -
-
2Ca 2Cb
2Cz
+
+ +
Vin Gm4
- -
-
Fig.11.5-075
2Cz
∑
gmi 2 80/1 80/1 80/1 80/1
v n2 =2vni2 ∑ = 2g eni
2
M12 M10 M11 M13
m1
i=1,4,10,12 VB2 M8 M9 VB2
i=1,4,10,12 100/1 100/1
where - Vout
+
vni2 = Noise power of transistors M1, M4, M10, and M12 VB3
10/1 10/1
VB3
VC
Using the definition of flicker-noise spectral density we get Vinp M6 M7
Vinn
40/1 40/1
K F ∆f M3
eni2f = S(f) = C WL f VB4
M1
60/1
M2
VB4
ox 60/1 60/1
Substituting into the above equation gives M4 M5
v n2 = Fig.11.6-08
∑
gmi 2 KFi ∆f 2
2g C W L f + 4kT 3g ∆f
m1 ox i i mi
i=1,4,10,12
It can be shown that the equivalent input-referred noise is minimized in the bandwidth of interest if
Comments:
• Uses low-Q, on-chip spiral inductors
• Linearity and noise is determined by the transconductors and the impedance level
Transfer function:
rs g mz 1 − g mz r ω 2
s 2 + s( − )+( ) s2 + z s + ω z
I out Ls C Ls C Qz
Gm ( s ) = (s) = g m = gm
Vin r 1 ω
s2 + s s + s2 + p s + ω p
2
Ls Ls C Qp
where 1 − g mz rs ωz
ωz = Qz =
Ls C rs g mz
−
Ls C
VBias for gm M3
100/2
VTune for gmz
Fig.11.6-09
Comments:
• Simple source-coupled differential pair is used for less noise
• Minimum size transistors is selected to minimize parasitic capacitance
• Linearity is a function of the gate-source overdrive voltage and hence the tail current
Q-Enhanced LC Filters
Second-order, Bandpass Filter:
+ -
G
Vin + M3 +
C L I2 +
- GM2 Center GM1 Vout
I3
RL - Freq. -
+
Filter -
Gain I1 Filter
R1 Gain
1GHz capability in bipolar.
L1A L1B
C1
+
+ Vout
Vin -
- M1A M1B
M2A M2B M3A M3B
CSA CSB
Ii Iq Io
Fig. HF07
†
W.B. Kuhn, “Design of Integrated, Low Power, Radio Receivers in BiCMOS Technologies, PhD Dissertation, Virginia Polytechnic Institute and State
University, Dec. 1995.
Example 5-2 - Fifth-order, Low Pass, OTA-C Filter using the Ladder Approach
Design a ladder, OTA-C realization for a Chebyshev filter approximation to the filter specifications
of TBP = -1dB, TSB = -25dB, fPB = 100kHz and fSB = 150 kHz. Give a schematic and component value for
the realization. Also simulate the realization and compare to an ideal realization. Adjust your design so that
it does not suffer the -6dB loss in the pass band. (Note that this example should be identical with Ex. 5-1.)
Solution
From Ex. 5-1, we know that a 5th-order, Chebyshev approximation will satisfy the specification. The
corresponding low pass, RLC prototype filter is
L5n =2.1349 H L3n =3.0009 H L1n =2.1349 H
+ 1ž +
C4n = C2n =
Vin (sn ) 1ž Vout (sn )
1.0911 F 1.0911 F
- -
Next, we must find the state equations and express them in the form of an integrator. Fortunately, the
above results can be directly used in this example.
’ 1 ‘ V1 ' -
V1(sn) = C s Gm11nVin(sn) - Gm21nV2(sn) - Gm31nV1(sn) (2) C1
1n n Gm31
+
Choosing L1n = C1n = 2.1349F gives Gm11n = Gm21n = Gm31n = 1S Fig.11.6-10
assuming that R0n = R’ = 1Ω. Also, double the value of Gm11n (Gm11n = 2S) in order to gain 6dB and
remove the -6dB of the RLC prototype.
1
C2n: V2(sn) = s R'C [V1'(sn) - V3'(sn) ] (3) V1'
+ V2
n 2n
Gm12
This equation can be realized by the OTA-C integrator shown which has one -
noninverting input and one inverting input. As before we write that V3' - C2
Gm22
1 ‘ ‘ +
V2(sn) = s C Gm12nV1 (sn) - Gm22nV3(sn) (4) Fig.11.6-11
n 2n
Choosing C2n = 1.0911F gives Gm12n = Gm22n = 1S
Fig.11.6-15
-10
-20
-30
dB
-40
-50
-60
10 100 1K 10K 100K 1M
Frequency (Hz) Fig.11.6-16
Vin Vout
+ - + - + -
- + - + - +
+ - + - -+
- +
+ - + - + -
- + - +
+ - + - + -
One section
V+ V-
(Io)
+ -
+Iout
V+ +
(Io)
V- -
-Iout
2C2 2C2 2C1 2C1
+ - V+ V- V+ V- + - + - V+ V-
(Io) (Io) (Io) (Io) (Io) (Io)
V+ V- + - + - V+ V- V+ V- + -
+Iin
-Iin
Characteristic Value
Frequency tuning range 50MHz - 130MHz
Integrator bias current 220µA @ fo = 130MHz
Integrating capacitors 8pF and 32pF
Power consumption with VCC = 5V 233mW @ fo = 130MHz
Quality factor ≈
3rd-order intermodulation distortion -45.6dB @ fo = 83MHz
Output current 3rd-order -14.5dBm @ fo = 83MHz
intercept point
Output noise power density @ fo = -152.4 dBm/Hz
83MHz
Frequency Tuning:
Main Filter
Vc
Phase
VCF External
Comparator
Signal
Tuning Methods
Indirect method:
• The filter is tuned in place (“in situ”). A master filter which is not in the system is tuned and the tuning
signals are applied to the slave filter which is in the system.
• This can be done at a high rate so the filter is constantly being tuned or infrequently such as at power up
or during some predetermined calibration period.
Direct method:
• The filter is taken out of the circuit and tuned. If another filter has been tuned it can be inserted in the
circuit while the other filter is being tuned.
Comments:
• Filter (slave) does not need to be disconnected from the system
• Two filters are required
Direct Tuning
Example:
Tuning procedure:
1.) The filter is take apart into several first- or second-order sections.
2.) Each section is tuned to the center frequency.
Comments:
• Necessary switches may influence the filter performance
Tuning procedure:
1.) The filter is removed from the circuit and a step voltage applied.
2.) The number of cycles in the ringing waveform and their period is used to tune the filter.
3.) Most tuning algorithms work better with a linear frequency tuning voltage relationship.
†
H. Yamazki, K. Oishi,, and K. Gotoh, , “An Accurate Center Frequency Tuning Scheme for 450KHz CMOS Gm-C Bandpass Filter,” IEEE. J. of
Solid-State Circuits, vol. 34, no. 12, pp. 1691-1697, Dec. 1999.
Oscillations between two states of tuning iteration can occur. This becomes a problem as the frequency
closely approaches the desired frequency.
Operation:
1.) A coarse tuning cycle using the LDC algorithm is used to find the actual frequency.
2.) The difference between the actual frequency and desired frequency is used to begin a successive
approximation cycle to fine-tune the filter.
Rupper||RM3(gmrds)2
Note: Amplitude = gm1Ro ≈ RM3 ≈ (gmrds)2 if Rupper >> RM3(gmrds)2
Constant-Q Tuning
By taking the advantage of the OTA topology and proper sizing, the Q-factor of the filter will remain
constant during tuning as illustrated below.
Tuning Algorithm
Since the poles of the VCO and zeros of the filter are identical, tuning the poles tunes the zeros.
Need to tune both the frequency and the notch depth.
Root locations:
Amplitude-Locked Loop
Tuning Response
Q tuning voltage →
VCO output
Frequency response →
SECTION 7 - SUMMARY
Comparison of Filter Types