Research Presentation #2 Academic Year 2012--2013 Insertion Loss Characteristics Of Commonly Used Passive Filters John Choma Ming Hsieh Department of Electrical Engineering Universityy of Southern California Los Angeles, California 90089-0271 (213) 740-4692 [Office] (818) 384-1552 [Cell] jjohnc@usc.edu @ [[E-Mail]] www.jcatsc.com [Course Notes & Technical Reports] Overview Of Presentation Two Port Filter Network Filter Metrics Power Loss Ratio Insertion I ti Loss L Passband Focus Lowpass Bandpass Filter Networks Butterworth Chebyshev Bessel-Thomson Research #2-- 2012-2013 Inductor Quality Factor Loss Implications Filter Pre-Distortion Engineering Assessment Future Work - Sensitivity Models Insertion Loss Of Lossless Filters 2 Insertion Loss Definition Zout(s) Signal S g a Power o e Att Load oad Port ot Load Power If Signal Is Coupled Directly To Load Represent This Power As Poo Expression 2 Voo Rl Vs 2 Rs Vi Pi Linear Two-Port Network Filter Rl Po 2 Vs Poo Rl R R Rl s l Load Power With Filter Inserted As Shown Represent This Load Power As Po Expression Vo 2 Rs Voo Vs 2 Poo Filter Vo 2 Vs Bypassed Po H(jω) Insertion se t o Loss oss Metrics et cs Rl Rl Power Loss 2 2 P V R 1 H(jω) l Ratio, plr p oo oo lr 2 P V R R o o s H(jω) l (j ) Insertion Loss,, IL, Is Decibel Value Of plr IL 10 log10 plr 20 log10 Voo Research #2-- 2012-2013 Insertion Loss Of Lossless Filters Rl Vo Vs Vo 3 Comments On Insertion/Power Loss 2 Rl R R s l 2 Power o e Loss/Insertion oss/ se t o Loss oss Comments 2 H(jω) IL Measures Losses Voo Incurred By Insertion Of IL 10 log l 10 plr 20 log l 10 Filter Between I/O Ports Vo Losses Are Aggravated By plr Poo V oo Pi Vo 1 Lossy Branch Elements Embedded In Filter Architecture Lowpass Filters Losses Are Aggravated Outside Passband Where Transfer Function Falls, Possibly Dramatically, As A Function Of Signal Frequency; These “Losses” Are Desirable, For They Help To Highlight Response Nature Of Stopband Lossless, Lowpass, Passive Filters Have H(0) = Rl /(Rl + Rs) Power Loss Ratio Is Inverse Of Filter Transfer Function Normalized To Zero Frequency Value Of Gain 2 plr Is Even Function Of Signal Frequency Poles And Zeros Poles Of Transfer Function Relate To Zeros off Power P Loss L R ti Ratio Zeros Of Transfer Function Relate To Poles of Power Loss Ratio Research #2-- 2012-2013 Insertion Loss Of Lossless Filters Rl plr R R s l 1 H(jω) 2 1 H(jω) H(0) 2 4 Power Loss Ratio For Lossless Filter Filter te Abstraction bst act o Average Incident Input Signal Power Is Pio Maximum Available Zout(s) Rs Vi Pio Lossless, Linear Two-Port Two Port Network Filter Vo Rl Pi Po Vs Power From Signal Pr Source Vs Actual Average Input Signal Power At Frequency ω If Rs = Zin(jω) [Conjugate Match Between Source And Input Impedances At Frequency ω] 2 Average Reflected Input Signal Power Is Pr Pr ρi (jω) Pio i ρi(ω) Is Input Port Reflection Coefficient Measured Z (jω) Rs With Respect To A Reference Resistance Of Rs ρi (jω) in Observed Average Input Signal Power Is Pi Zin (jω) Rs Observed Average Output Signal Power Is Po Pi Pio Pr Po = Pi Because Of Lossless Filter Architecture 2 1 ρi (jω) Pio Power Loss Ratio, plr Measures Power Reduction Due To P 1 p io Impedance Mismatch lr 2 Po 1 ρ (jω) i Unity Loss Ratio (No Loss) When Zin(jω) = Rs Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 5 Properties Of The Reflection Coefficient Z (jω) (j ) Normalized o a ed Input put Impedance peda ce zn (jω) (j ) in rn ((ω)) jx j n (ω) ( ) Rs rn(ω) = rn(−ω) For Physically Realizable Driving Point Input Resistance (Even Frequency Function) xn((−ω) ω) = −x xn(ω) For Physically Realizable Driving Point Input Reactance (Odd Frequency Function) Z (jω) Rs r (ω) 1 jxn (ω) Reflection Coefficient ρi (jω) in n Zin rn (ω) 1 jxn (ω) Magnitude Squared i (jω) Rs r (ω) 1 jxn (ω) rn (ω) 1 jxn (ω) 2 ρi (jω) ρi (jω)ρi (-jω) n rn (ω) 1 jxn (ω) rn (ω) 1 jxn (ω) 2 Alternative Expression M (ω) rn (ω) 1 xn2 (ω) r (ω) 1 jxn (ω) rn (ω) 1 jxn (ω) M (ω) 2 ρi (jω) n j n (ω) ( ) rn ((ω)) 1 jx j n (ω) ( ) M (ω) ( ) 4rn (ω) ( ) rn ((ω)) 1 jx Power Loss Ratio In Terms Of Reflection/Impedance Parameters p R h #2 2012 2013 2 rn (ω) 1 xn2 (ω) M (ω) 1 1 2 4rn (ω) ( ) 4rn ((ω)) 1 ρ(jω) 1 l lr I ti L Of L l Filt 6 Example: Lowpass Butterworth Filter 2 nth Order Lowpass Transfer Relationship H(jω) Maximally Flat Magnitude Butterworth H(0) Frequency Response 1 B Designates 3-dB Bandwidth Of Filter H(0) Represents Zero Frequency Value Of Transfer Function ω B 2n Power Loss Ratio (plr) Of nth Order, Lowpass Butterworth Filter ω 1 plr 2 B H(jω) H(0) 1 1 2n Insertion Loss (IL) Of nth Order, Lowpass Butterworth Filter 2n ω IL 10 log10 plr 10 log10 1 B Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 7 Insertion Loss Frequency Response (LP) Lowpass Butterworth Filter 80 Insertio on Loss In D Decibels 70 60 50 40 30 20 10 0 0.10 0.16 0.25 0.40 0.63 1.00 1.58 2.51 3.98 6.31 10.00 N Normalized li d Signal Si l Frequency, F ( /B) (ω/B) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 8 Expanded Insertion Loss Response (LP) Lowpass Butterworth Filter 3 Insertio on Loss In Decibels 25 2.5 2 1.5 1 0.5 0 0.10 0.16 0.25 0.40 0.63 1.00 Normalized Signal g Frequency, q y, ((ω/B)) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 9 Comments LP On Insertion Loss Responses General Comments Response Plots Applicable Only To Lowpass Butterworth Filters Frequency Scale Is Normalized To 3-dB Bandwidth Of Filters Design D i Ob Observations ti Insertion Losses Are Less Than 3 Decibels Within Filter Passband, Independent of Filter Order Passband P b d Extends E t d From F Zero Z Frequency F -ToT 3-dB 3 dB Bandwidth, B d idth B Insertion Losses Dramatically Changes With Order n Outside Passband (Which Is Termed The Stopband) To Be Expected In That Large n Encourages Rapid Frequency Response Rolloff In Stopband Insertion Loss Approaches 20n dB For High Signal Frequencies Insertion Losses Are Reduced For Increased Filter Order Within Filter Passband Higher Order Implies Filter Frequency Response Is Flatter Over Wider Passband Frequencies Very V High Hi h O Order d L Lowpass B Butterworth tt th Filt Filter A Approximates i t Id Idealized li d “Brick “B i k Wall” Filter Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 10 Example: Butterworth Bandpass Filter s Lowpass o pass To o Bandpass a dpass Transformation a s o at o ω c p BQ p Is Complex Frequency For Lowpass Prototype cω s c B Represents Bandwidth Of Lowpass Prototype ω ωc Is Tuned Center Frequency Of Bandpass Realization Bc c Q Qc Is Quality Factor Of Bandpass Realization c Bc Represents 3-dB Bandwidth Of Bandpass Filter Transformed Transfer Function H(jω) H(0) 2 1 2n Lowpass p To Bandpass H(jω) H(jωc ) 2 1 ω ωc ω 1 Qc 1 ω ω B c Insertion Loss (IL) Of Bandpass Butterworth Filter ω ωc 1 Qc plr 2 ω ω H(jω) H(jωc ) c 1 IL 10 log10 plr 10 log10 1 Research #2-- 2012-2013 2n ω ωc Qc ω ω c Insertion Loss Of Lossless Filters 2n 2n 11 Insertion Loss Frequency Response (BP) Bandpass Butterworth Filter--Constant Filter Constant Q 140 Qc = 5 Insertion n Loss In De ecibels 120 100 80 60 40 20 0 0.10 0.16 0.25 0.40 0.63 1.00 1.58 2.51 3.98 6.31 10.00 Normalized Signal Frequency, (ω/ωc ) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 12 Expanded Insertion Loss Response (BP) Bandpass Butterworth Filter--Constant Filter Constant n 14 n= 1 Insertion Loss In Dec cibels 12 10 8 6 4 2 0 0.75 0.80 0.85 0.90 0.95 1.00 1.05 1.10 1.15 1.20 1.25 Normalized Signal Frequency, (ω/ωc ) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 13 Comments On BP Insertion Loss Responses Power o e Loss oss Ratio at o Neighborhood Of Tuned Center Frequency ω = ωc + Δω ω ωc ω ωc 2Δω 1 Δω 2Δω , Δω 2ω ω ωc c ωc ω ωωc ωc 2 c 2ω ωc 2n 2n ω ωc 2Qc Δω Q 1 c ω ω ω c c 10 IL 10 log10 plr ln plr 4.343 ln plr Resultant ln 10 2 2n 22n I Insertion ti 2Qc Δω 2n Δω Loss IL 4.343 ln 1 4.343 2Q c ω ω c c Observations No Insertion Loss At, And Very Near, Tuned Center C Frequency, Which Implies Unity Center Frequency Gain Insertion Loss Increases As (2n)th Power Of Twice The Filter Quality F t Factor Resultant Power Loss plr 1 Ratio Δω ωc Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 14 Lowpass Chebyshev Filter 2 Squa ed Transfer Squared a s e Relationship e at o s p H(jω) 1 ε Defines Allowable Response Ripple Within H( 0 ) 1 ε 2Cn2 (ω) Filter Passband Ripple (r) In Decibels r 20 log g10 1 ε 2 10 log g10 1 ε 2 Filter Passband Extends To Radial Frequency Of ωp Passband Frequency Is Generally Smaller Than Filter 3-dB Bandwidth Special Case Is That ωp Is Exactly The 3 3-dB dB Bandwidth If r = 3 dB Response In Passband At Zero Frequency: At Passband Edge: Chebyshev Polynomial, Cn(ω) Squared Sinusoidal Cn (ω) Cn2 (ω p ) 1 B Bracketed k dR Response Wi Within hi Passband P b d Research #2-- 2012-2013 cos n cos 1 ω ω p , 0 ω ω p 1 coshh n cosh h 1 ω ω p , ω ω p 1 1 , n even Cn2 (0) 0 , n odd Insertion Loss Of Lossless Filters 1 1 ε2 H(jω) (j ) 1 H( 0 ) 15 Comments On Chebyshev Response H(jω) H( 0 ) General Ge e a Transfer a s e Function u ct o Within Filter Passband, 0 ≤ ω/ωp ≤ 1 1 ε 2Cn2 (ω) Frequency Response Modulates Because 1 H(jω) 1 Of Sinusoidal Nature Of Chebyshev H( 0 ) 1 ε2 0 ≤ ω/ωp ≤ 1 Polynomial H(jω p ) At Passband Edge, Frequency Response Is 1 Down By An Amount Related To Ripple H( 0 ) 1 ε2 H(jω) 1 H( 0 ) 1 ε 2 cos 2 n cos 1 ω ω p r ripple 10 log10 1 ε 2 Outside Filter Passband, ω/ωp ≥ 1 Frequency Response Attenuates Sharply Because Of Hyperbolic Nature Of Chebyshev y Characteristic Polynomial y Response Outside Passband Is Continuous With Passband Response At Passband H(jω) 1 Edge g ((ω = ωp) H( 0 ) 1 ε 2 cosh 2 n cosh 1 ω ω p 1 Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 16 Example: Lowpass Chebyshev Filter Power o e Loss oss Ratio at o plr 1 2 1 ε 2 cos 2 n cos 1 ω ω p 0 ω ω p 1 1 ε 2 cosh 2 n cosh 1 ω ω p ω ω p 1 Insertion Loss IL 10 log10 plr r ripple 10 log10 1 ε 2 Plots (Next Two Slides) Stopband (ω > ωp) Increased Insertion Loss With Increasing Filter Order Larger Rate Of Insertion Loss Increase Than With Butterworth Filter Passband (ω ≤ ωp) Insertion Loss Exacerbated By Response Ripple Peak Passband Insertion Loss Identical To Filter Ripple Insertion Loss Identical To Response Ripple At Passband Edge Filter Order H(jω) H(0) For Odd Order Filter, Insertion Loss Approaches Zero Decibels At Very Low Signal Frequencies For Even Order Filter Filter, Insertion Loss Approaches Response Ripple At Very Low Signal Frequencies Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 17 Insertion Loss Frequency Response (LP) Lowpass Ripple Chebyshev Filter--Constant Ripple 120 ripple = 2 dB Insertion n Loss In De ecibels 100 80 60 40 20 0 0 10 0.10 0 16 0.16 0 25 0.25 0 40 0.40 0 63 0.63 1 00 1.00 1 58 1.58 2 51 2.51 3 98 3.98 6 31 6.31 10 00 10.00 Normalized Signal Frequency, (ω/ωp ) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 18 Passband Insertion Loss Response (LP) Lowpass Chebyshev Filter--Constant Filter Constant n Insertion L Loss In Deciibels 2.0 n=3 1.5 1.0 0.5 0.0 0.10 0.13 0.16 0.20 0.25 0.32 0.40 0.50 0.63 0.79 1.00 Normalized Signal Frequency, (ω/ωp ) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 19 Chebyshev Bandpass Filter Frequency Transformation s ω B ωc ω p Q c c Q cω c s p c p Power Loss Ratio And Insertion Loss r 10 log10 1 ε 2 ω ωc 2 2 1 1 ε cos n cos Qc passband ωc ω plr ω ωc 2 2 1 1 ε cosh n cosh Qc stopband IL 10 log10 plr ωc ω Insertion Loss Plots Metrics Fourth Order (n = 4) Quality Factor (Qc = 7) Ripple (r = 0.25 dB, 0.50 dB, 1 dB) Plot On Next Slide Comments Subsequent To Next Slide Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 20 Passband Chebyshev Insertion Loss (BP) Bandpass Chebyshev Filter Filter--Constant Constant n & Qc Insertion Loss In Dec cibels 2.0 n=4 Qc = 7 1.5 1.0 0.5 0.0 0.93 0.95 0.97 0.99 1.01 1.03 1.05 1.07 Normalized Signal Frequency, (ω/ωc ) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 21 Comments On Chebyshev Insertion Loss Passband assba d Response espo se Metrics Shown For 4th Order Bandpass Filter Shown For Quality Factor, Qc, Of 7 Radial Center Frequency Is ωc Plotted Only For Passband Response Passband Width Is ωp, Which Is Not 3 dB Bandwidth, 3-dB Bandwidth B 3-dB Bandwidth, B Insertion n Loss In Decibels 2.0 n=4 Qc = 7 1.5 1.0 0.5 0.0 0.93 0.95 0.97 1.01 1.03 1.05 1 1 cosh cosh 1 n r 10 p 10 1 B Comments Insertion Loss Is Sensitive To Passband Ripple Maximum Insertion Loss In Passband Is The Ripple Maximum Insertion Loss 0.99 1.07 N Normalized li d Signal Si l Frequency, F ( / c) (ω/ω Occurs At Tuned Center Frequency For Even Integer n Occurs At Band Edges For Odd Integer n Small Ripple Case Chebyshev Ch b h IL Less L D Dramatic ti Than Th Butterworth B tt th Filter Filt IL Within Withi P Passband b d Insertion Loss Exceeds That Of Butterworth In Neighborhood Of Tuned Center Frequency Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 22 Chebyshev 3-dB Bandwidth 33-dB dB Bandwidth Converges To Passband Bandwidth For Large Filter Order Norm malized 3-dB B Bandwidth. (B/ωp ) 2.00 1.50 5th Order (n = 5) 1.00 0.10 0.40 0.70 1.00 1.30 1.60 1.90 2.20 2.50 2.80 Passband Ripple (r) In Decibels Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 23 Lowpass Bessel-Thomson Filter () 1 Normalized o a ed H(s) 2 3 4 Transfer H(0) 1 sTdo a2 sTdo a3 sTdo a4 sTdo Function Groupp Delayy At Zero Signal g Frequency q y Is Tdo Coefficients For nth Order Filter 2 n 1 n 2 n 1 a2 a3 2n 1 3 2n 1 2n 2 n 1 n 2 n 3 3 2n 1 2n 2 2n 3 2 m 2n m ! n! a4 am m! 2n ! n m ! Squared Fourth Order Transfer Function Relationships H(jω ) H(0) plr Research #2-- 2012-2013 2 1 2 1 a ωT 2 a ωT 4 ωT 2 1 a ωT 2 2 do 4 do do 3 do 1 H(jω) H(0) 2 2 IL 10 log10 plr Insertion Loss Of Lossless Filters 24 Lowpass Bessel Filter Insertion Loss Lowpass Bessel Bessel-Thomson Thomson Delay Equalizer 40 Insertion Loss In De ecibels 35 30 Order = 4 (n = 4) 25 Order = 3 (n = 3) 20 Order = 2 (n = 2) 15 10 5 0 0.10 0.16 0.25 0.40 0.63 1.00 1.58 2.51 3.98 6.31 10.00 Normalized Signal Frequency, (ωTdo ) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 25 Comments On Bessel Thomson Performance Bessel Thomson Filter Commonly Known As Bessel Thomson Delay Equalizer Provides Maximally Flat Group Delay Over Filter Passband Amount Of Maximally Flat Delay Is Zero Frequency Delay Delay, Tdo Insertion Loss Characteristics Minimal Insertion Loss Within Filter Passband; Slightly Better Than B tt Butterworth th Filter Filt Insertion Loss Within Filter Passband Actually Improves Slightly With Filter Order 3-dB Bandwidth Daunting Challenge To Quantify In Closed Form 3-dB Bandwidth Estimates, B n = 2 B = 0.7913/Tdo = 2.374a2 /Tdo n = 3 B = 0.7855/Tdo = 1.960a2 /Tdo n = 4 B = 0.7855/Tdo = 1.833a2 /Tdo Bandpass Insertion Loss Follows Insertion Footprints Of Butterworth and Chebyshev Filters Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 26 Example: Lowpass Impedance Converter L Circuit C cu t Sc Schematic e at c Diagram ag a Filter Metrics R Self Resonant Frequency Is ωo C R Circuit Quality Factor Is Q Transfer Function If V FILTER RL Filter Is Removed, KL: K L RL RS Static Effect Of RS RL Inductor Quality KR RL RS RL R Factor, KR: H(0) KL KR RS RL R Zero Frequency Gain Is H(0) K z RS RL 1 Impedance Conversion Factor Is Kz: H(0) KL KR I/O Transfer Function H(s) Vo 2 2 Vs Self-Resonant Self Resonant s s s s 1 1 Frequency Qωo ωo Qωo ωo R Vi Vo S L S ωo 1 H(0)K z LC Circuit Quality Factor Derives From: Research #2-- 2012-2013 1 1 Q RL Q = 1 Insertion Loss Of Lossless Filters L H(0) C RL R H(0)K z C Kz L 2 for maximally flat magnitude 27 Insertion Loss Analysis Power o e Loss oss Ratio at o Voo /Vs Is Transfer Function With Signal Source Coupled Directly To Load Vo /Vs Is Transfer Function With Filter Inserted Between Source And Load Terminations V H(jω ) o Vs KL KR ω jω 1 ω Qωo o Resultant Power Loss Ratio Resultant Insertion Loss Research #2-- 2012-2013 2 p lr H(jω ) p Voo Vs 2 lr 2 Voo Vo 2 Voo Vs Vo Vs 2 RL 2 KL R R S L 2 K L K R 2 2 2 2 1 ω ω Qωo Q ωo 2 4 1 ω ω 1 2 2 2 ω ω Q o V V o oo s 2 Vo Vs KR V IL 10 log10 plr 20 log10 oo Vs Insertion Loss Of Lossless Filters Vo Vs 28 Converter Insertion Loss Insertion Loss In Decibels 40 KR = 0.9 26 Maximally Flat Magnitude 12 -2 0.10 0.16 0.25 0.40 0.63 1.00 1.58 2.51 3.98 6.31 10.00 Normalized Signal Frequency, (ω/ωo ) Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 29 Comments On Converter Insertion Loss Insertion se t o Loss oss Plot ot General Observations Low Frequency Insertion Loss Determined Largely By Metric KR KR Is Intimately Related To Resistance R, Which Accounts For Finite Inductor Quality Factor KR = 1 For Ideal Inductor (R ( = 0)) Insertion Los ss In Decibels 40 KR = 0.9 26 Maximally Flat Magnitude 12 -2 0.10 0.16 0.25 0.40 0.63 1.00 1.58 2.51 3.98 6.31 10.00 Normalized Signal Frequency, (ω/ωo ) Maximally Flat Magnitude Case Gives Small, Monotonically Increasing Passband Insertion Loss L Large Ci Circuit it Q Corresponds to Small Second Order Damping Factor Yields Magnitude Response Peaking In Passband Correspondingly delivers Non-Monotonic Non Monotonic Insertion Loss Response Small Circuit Q Corresponds To Second Order Overdamping Gives Monotonically Increasing Insertion Loss Response Because Frequency Response Is Characteristically Monotonic Terrible Passband Insertion Loss Response Due To A Passband That Is Constrained By Small Bandwidth Arising From Overdamping Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 30 Modeling And Metrics Of Inductance Loss Ideal dea Inductance ducta ce Simple Inductive Impedance No Winding Or Other Losses I(s) V(s) I(s) Z(s) sL Practical Inductance Choose A Normalizing Frequency, ωα Can Be Bandwidth Of Lowpass Filter Or I(s) () L L V(s) RLL V(s) V(s) I(s) Z(s) sL RLL q y Of Bandpass p Filter Center Frequency Can Be Self-Resonant Frequency Of Filter Network Choose A Normalizing Impedance, Zα Can Be Signal Source or Terminating Load Resistance Of Network Can C B Be Ch Characteristic i i IImpedance d Of C Considered id d S Structure Normalized Frequency, p: p s ωα ωα L Q L Quality Factor, QL, Of Inductor RLL Normalizing Circuit Parameters Inductance, Lα: Lα Zα ωα Capacitance, Cα: Cα 1 ωα Zα Normalized Variables (Impedance Zn, Inductance Ln, Capacitance Cn, Frequency, p) Z n Z Z α Ln L Lα Cn C Cα p s ωα Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 31 Inductance Loss Mathematical Model L RLL Practical act ca Inductance ducta ce Analysis Of Net Inductor Impedance V(s) Impedance s RLL s ωα L ωα L RLL ω ω L ω sL RLL α α α Z nL (s) Zα Zα Zα Normalized Impedance Of L 1 1 Z (p) p p Ln nL Practical Inductor Lα QL QL Approximation pp “Practical” I(s) Inductor Invokes Only Frequency Invariant, Constant Series Resistance Analytical Procedure Replace “Practical” Inductance With Normalized Inductance, Ln = L/Lα Write W it Conventional C ti l Ki Kirchhoff hh ff E Equilibrium ilib i E Equations, ti B Butt U Using i a complex l Frequency Variable In 1 s 1 s RLL Inductor Impedance Of: p Q ω Q ω ω L Invoke This Procedure For L L α α α E h IInductance Each d IIn Network N kU Undergoing d i S Scrutiny i Procedure Gives Transfer Function With All Inductance Qs Considered Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 32 Modeling And Metrics Of Capacitance Loss Ideall Capacitance Id C it Simple Capacitive Admittance No Dielectric Or Other Losses I(s) V(s) Y(s) sC I( ) I(s) C V(s) C Practical Capacitance I(s) R Choose A Normalizing Frequency, ωα, And A Normalizing Admittance, Yα, As In Inductor Case V(s) Normalized Frequency, p: p s ωα I(s) V(s) Y(s) sC 1 R CC Quality Factor, QC, Of Capacitor QC ωα RCC C Capacitive Quality Factor Generally Much Larger Than CC Inductor Quality Factor Capacitors Are Far Less Lossy Than Are Inductors Normalizing g Circuit Parameters Inductance, Lα: Lα Zα ωα Capacitance, Cα: Cα 1 ωα Zα Normalized Variables (Admittance Yn, Inductance Ln, Capacitance Cn, Frequency, p) Yn Y Yα Z αY Ln L Lα Cn C Cα p s ωα Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 33 Capacitance Loss Mathematical Model C Practical act ca Capac Capacitor to I(s) RCC Analysis Of Net Capacitor Admittance V(s) Normalized Admittance s 1 1 ω C sC α s ω RCC RCC 1 α YnC (s) ωα Z αC Yα Yα ω ω R C α CC α Normalized Admittance Of C 1 1 Y (s) p p Cn nC Practical Capacitor Cα QC QC Approximation Of “Practical” Practical Capacitor Invokes Only Frequency Invariant, Constant Shunt Resistance Analytical Procedure Replace p “Practical” Capacitance p With Normalized Capacitance, p , Cn = C/Cα Write Conventional Kirchhoff Equilibrium Equations, But Using a complex Frequency Variable In Capacitor Admittance 1 s 1 Of: p Q Invoke This Procedure For Each Capacitance C ωα QC In Network Undergoing Scrutiny Procedure Gives Transfer Function With All Capacitance Qs Considered Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 34 Approximate System Level Analysis First st O Order de Analysis a ys s And d Design es g Replace Normalized Complex Frequency By s 1 p Q-Corrected Normalized Frequency: ωα Q L Leads To Conservative Estimate Of Effect Of Losses Quality Factors Of Capacitors Are Relatively Unimportant Because They Are Generally Much Larger Than Inductive Counterparts Pre-distortion: Close Examination Of Lossy Result Leads To Guidelines For Modifying Modif ing Val Values es Of Energ Energy Storage And Loss Lossy Branch Elements Example Return To Lowpass Impedance Converter Considered Earlier Execute Analysis With Utilized Inductance Presumed Ideal Resistance R Is Set To Zero (No Loss In Circuit Inductance) Let Resultant Self-Resonant Frequency Be Designated As ωx Let Resultant Circuit Quality Factor Be Designated As Qx Analysis: Normalized Frequency Is (s/ωx) Replace (s/ωx) By (p + 1/QL) Implication Of This Simplifying Approximation Is That All Energy Storage Elements Exude Identical Qs Result Is Approximate (Potentially Significantly Conservative) Result Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 35 Filter Analysis L Filter te With t Ideal dea Inductance ducta ce I/O Transfer Relationship, H(s) R Vo H(0) H(s) 2 Vs C R s s M ti Metrics LOSSLESS 1 V FILTER Qx ωx ωx Self Resonant Frequency, ωx RL RS 1 Quality ωx RS LC H(0)K z LC Factor Normalized Frequency Is 1 1 L H(0) ( ) C (s/ωx) RL H(0)K z Qx RL C K z L Replace By (p + 1/QL) Large QL Presumption RS LC C RL p 1 Q 2 p 2 2 p Q RS RL RS L RL RS L L Comments ωx Is A Measure Of 3-dB Bandwidth Of Filter 1/Qxωx Measures The Group I/O Delay Of Filter At Low Frequencies Qx Is A Measure Of Relative Stability And Settling Time Of Filter S Vo R=0 Vi L S • Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 36 Approximate Impact Of Inductor Q Revised e sed Transfer a s e H(s) Relationship 1 (Approximate) Alternative Form H(0) H(0) 2 p 1 QL 2 2p s s 1 p Qx QL Qx ωx ωx H(0) 1 1 Qx QL H(0) H(p) p 1 QL 2Qx 2 2p 1 p 1 Qx QL p QL p2 1 1 1 Qx 1 1 1 ω L ΓL 1 x Qx QL Qx QL Qx QL QL R Comments Infinitely Large QL Reduces Alternative Transfer Function Form To Normalized Form Of Original Network Transfer Function Resultant Self-Resonant Frequency q y Increases By y Square q Root Of ΓL Quality Factor (Qx) Increases If 2Qx2 > 1 Resultant Zero Frequency Gain Reduced By ΓL Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 37 Comments On Inductor Loss Effects 1 General Ge e a Co Comments e ts Γ 1 ωx L 1 L Q Qx QL L Procedure R Assumes That All Energy Storage Elements Exude The Same Quality y Factor Capacitors Have Larger Qs Mitigation Increase Calculated QL By Nominally A Factor Of Two Simulate “DC” Transfer Function Compare To Calculated Value Of “DC” DC Transfer Extract Effective Value Of QL L R Vi Vo RS VS C RL FILTER H(0) ( ) ΓL H(p) p 1 Qx Zero Frequency Group Delay (Tdo) Increases If 2Qx2 > 1 And A d ΓL ≠ 1 1 2Q 2 Γ 1 p 2 x L ΓL ΓL 1 2Qx2 ΓL 1 Tdo ωx Qx ΓL Design “DC” Gain, H(0), Self-Resonant Frequency, ωx, & Circuit Qx Affected Quantification Of These Effects Leads To Optimization Via Branch Element “Pre-Distortion” Adjustments Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 38 Future Work Qua ty Factors Quality acto s O Of Energy e gy Sto Storage age Elements e e ts Examine Efficient, Realistic Means Of Considering Capacitive Losses Examine Effects Of Skin Effect and Self-Resonance In Inductances Pole Positions Discern Perturbation In Real And Imaginary Parts Of Poles As A Result Of Energy Storage Element Losses Generally, Generally Pole Locations Move Closer To Imaginary Axis In Complex Frequency Plane By an Amount Of 1/QL Circuit Models Development Of Models That Project Sensitivity To Energy Storage Losses Possible Use Of Network Adjoint Topologies Available To Ascertain Performance Sensitivity To Branch Element Parameters Already Embedded In Cadence Sceptre For Sensitivity Analysis Trick Is To Apply, Realistically, Adjoint Sensitivity Methods To Insertion Loss Frequency Response Research #2-- 2012-2013 Insertion Loss Of Lossless Filters 39