Electronics for Telecommunications

2 Selectivity, matching and filtering

RLC and crystal resonators y The maximum power transfer theorem M t hi circuits: L h Matching i it L-shaped, -shaped and T h d h d d T-shaped circuits d i it Filter properties and specifications in magnitude and phase Filter Types: Butterworth, Chebychev and Bessel g procedure: the Insertion Loss Method ( (ILM) ) General filter design p Component selection: filter order and tables Impedance denormalization and frequency transformations Passive integrated components: resistors, capacitors and inductors SAW filters basics
1

Overview
Electronics systems for telecommunications (especially transmitters and receivers) rely on many passive components (also integrated) Passive components are essential to design selective networks (i.e. analog filters) and matching circuits. At very high frequency distributed parameter circuits (i.e. transmission lines) have to be used rather than lumped components. Using distributed or lumped p p g p elements depends on the dimensions of the circuit considered. However, the design procedure is similar in both cases. Given any passive network provided with reactive elements we refer to the quality factor Q of the circuit at a frequency as:
Energy stored in the reactive elements

Q=

Ec Pd
Dissipated power 2

Parallel RLC resonator

Admittance: Y ( j ) =
1 1 1 1 1 j +j = j + jC R XL XC R L
f0 = 1 2 LC

(Anti) Resonance freq. (XL=Xc): 0L = 1 0C Q-factor at 0: Bandwidth:

Q = 0
1 CV 2 1 RI 2 2 2

= 0 RC =

R 0 L

B = 2f3dB

f 1 = 0 2CR Q

The higher Q, the narrower the bandwidth is for a given f0 |Z(f)|

f3dB f3dB

f0 The impedance is maximum in f0

3

Series RLC resonator

Impedance: Z ( j ) = R + jX L + jX C = R + jL j L C R Resonance frequency (XL=Xc): 0L = Q factor Q-factor of the resonator: Q = Bandwidth: B =
f0 R = Q 2L

1 C

1 0C

f0 =

1 2 LC

0 L 1 = R 0 RC

Similarly to the previous case, the higher Q, the narrower B for a given f0 |Z(f)|
f3dB f3dB

f0 The impedance is minimum in f0

4

Other RLC resonators

Any resonant circuit can be transformed into a standard parallel or series one in in a limited frequency range Example C LS C RS Assume that Qs=Qp=Q and that the behavior of the two resonator must be the same around 0 LP RP

Rs + j0 Ls =

j0 Lp R p R p + j0 Lp

2 Rp R p = Rs + 1 = Rs Q 2 + 1 L2 0 p

L Q= = 0 s 0 Lp Rs

Rp

Q 2 + 1 Lp = Ls Q2

These equations are absolutely general, i.e. they hold also in the case of a serial to parallel transformation and vice versa around 0

Crystal resonators 1
For f<50 MHz, plain discrete components (i.e. real R,L,C) are adequate to build resonators. If a very high Q is required crystal resonators are to be preferred Crystals resonators are p y piezoelectric slices ( (usually q y quartz) ) When an oscillating signal is applied to a crystal, it starts vibrating at a resonance frequency (50 kHz-50 MHz) depending on the size (thickness) of the crystal
fr =
fa = 2 L1

1 2 L1C1
1 C1C0 C0 + C1

Z (s ) =

s 3L1C1C0 + s 2C0C1R1 + s (C1 + C0 ) R 0 1 0

s 2L1C1 + sC1R1 + 1

Z ( j ) =

(1 L C )
2 1 1

CC j (C0 + C1 )1 2L1 1 0 C0 + C1
6

Crystal resonators 2
Usually C1<<C0 The crystal resonator bandwidth and the Q factor are:
fa f r = f r C1 << 1 2C0

Q=

s L1 1 = R1 sC1R1

When the crystal resonates around fr the approximate electric model is: C1 Series or resonant scheme L1 When the crystal resonates around fa the electric model is: L1
C1C0 C0 + C1
Parallel or anti-resonant scheme

The resonating mode excited by a crystal oscillator depends on the particular circuital scheme in which the resonator is inserted
7

High-frequency resonators
Microwave resonators for RF applications (100 MHz-100 GHz) can not be implemented using traditional techniques because: Traditional crystal resonators are too thin frail devices Circuit connections become transmission lines and lumped circuit components such as R, L and C are impossible to design exactly because of parasitic effects associated with connections, pins packages connections pins, packages

Solutions
Resonant cavities (e.g. transmission lines or waveguides which are shortcircuited at both ends) very used in high-end telecommunication equipment Surface Acoustic Wave (SAW) resonators (see later)

The maximum power transfer theorem

Zs=Rs+jXs ZL=RL+jXL vs(t)=Vscos(st) () ( )

To transfer the maximum amount of active power from a source with a given internal impedance to the load, the impedance of the load must be the complex conjugate of p j g the source.

g power p provided to the load is g given by: y Proof: the average active p 2 RL PL = VSeff (RS + RL )2 + ( X S + X L )2

The maximum of this quantity is achieved for: 1) 2)

XL = XS

dPL =0 dRL

RL = RS

PLmax =

VSeff 4RL

Maximum power M i that can be transferred to a load

Types of matching networks

Matching (i.e. ZI=ZL*) can be obtained inserting 2-port networks (e.g. T, or L networks) made up of lumped circuit elements before the load ZL
L-shaped networks Case (a): |ZL|>|ZI|

ZL ZI

jX1

downward conversion

Case (b): |ZL|<|ZI|

ZL jB2 ZL ZI

jX2 jB1

upward conversion

ZL

The main problem of L networks is that we have just 2 degrees of freedom, i.e. bandwidth and matching cannot be set independently. T and networks solve the previous problem because they have a further degree of freedom ZL ZI
-shaped network T-shape network

j jX2 jB1 jB2 ZL ZI

ZL

j jX1 jB2

j jX3 ZL
10

L-shaped impedance network design - 1

Upward transformation case (RL<RI) RI L2 C1
B1=j0C1 X2=j0L2

RL

RI

C1

Lp

Rp

R p = RL Q 2 + 1 Q 2 + 1 j0 Lp = j0 L2 Q2

Question: what are the values of C1 and L2 to assure a perfect matching at 0

1 1 ' YL = + j 0C1 0 LP RP 2

RI = R L Q + 1
X I = 0C1

Q=

RI 1 RL

1 =0 0 LP

L2 =

QRL

C1 =

1 Q 2 0 L2 + 1 2 Q
2

Notice that C1 and L2 are determined univocally, while Q is given by the impedance transformation ratio the bandwidth around 0 is fixed.
11

L-shaped impedance network design - 2 p p g

L2 RI Downward transformation case (RL>RI) RI L2 Rs Cs C1
B1=j0C1 X2=j0L2 j

RL

Rs =

RL Q2 + 1

Q 2 + 1 Cs = C1 Q2

Q Question: what are the values of C1 and L2 to assure a perfect matching at 0 p g

1 RP 2 1 + j 0C1 0 LP
RI = RL Q2 + 1
Q= RL 1 RI

' YL =

X I = 0CS

1 =0 0 L2

Q C1 = 0 RL

L2 =

1 Q 2 0 C1 + 1 2 Q
2

12

-shaped impedance network design

L RI C1 C2 RL RI L1 C1 RZ L2 C2 RL

QLeft QRight Question: what are the values of C1, L and C2 to assure a perfect matching at 0 within a given bandwidth B? In case of matching
R Z = RL RI = R Z
Q 2 + 1 1 =0 X Z 0C2 Q2 L 0 2 1 =0 X I 0C1 Q 2 + 1 0 L1 Q2
Q= f0 = QLeft + QRight = B RI RL 1 + 1 RZ RZ RI RL + RZ RZ

RZ

L2 =

1 2 2 QRight + 1 0 C2 2 QRight g

C2 =

QRight 0 RL

L1 =

QLeft R z

C1 =

1 2 2 QLeft + 1 0 L1 2 QLeft
13

Both bandwidth and matching can be obtained

T-shaped impedance network design

L1 RI C L2 RL RI L1 C1 QLeft RZ L2 C2 QRight

This is dual to the previous case

Q= f0 = QLeft + QRight = B RZ RZ 1 + 1 RI RL RZ RZ + RI RL

RZ

L2 =

QRight RL 0

C=

Q 0 R z

L1 =

QLeft RI 0
14

Example
RL= 50 RI = 5 0 = 2 1 GHz Bd=25 MHz

Design the L-shaped and -shaped matching networks centered in 0 meeting the wanted bandwidth constraint 1) L-shaped (downward transformer) Q=3 (fixed) C1=9.55 pF; L2=2.4 nH but B=333 MHz. The bandwidth specifications are not met. 2) -shaped from Bd we obtain that Q=40 Rz=0.054 QRight=30.4 and QLeft=9 6 =9.6 C1=305 pF; C2=96 8 pF; L=0.344 nH. The bandwidth =96.8 L=0 344 nH specifications are met, but the components can be hardly built using integrated technologies: capacitances are too large and inductances too small. small
15

Introduction to filters
Analog filters must provide good matching (i.e. power transfer) over a certain bandwidth, while assuring adequate selectivity. Filters can be of different types, i e Low-pass filter (LPF) High-pass filter (HPF) types i.e. Low pass (LPF), High pass (HPF), Bandpass filter (BPF), Bandstop filter (BSF). In each category filters can be further divided into active and passive. In category, passive active filter there can be amplification of the signal power in the passband region, whereas passive filter do not provide any amplification.

Ref. Pictures from H. Khorramabadi presentation, Analysis and Design of VLSI Analog-to-Digital Interface Integrated circuits, UC Berkeley, 2009 I t t d i it B k l

16

Filter frequency characteristics

Ref. Ref Pictures from H Khorramabadi presentation, Analysis and Design of VLSI Analog-to-Digital Interface Integrated H. presentation Analysis Analog to Digital circuits, UC Berkeley, 2009

17

Group delay and phase distortion - 1

18

Group delay and phase distortion - 2 p y p

19

An example of critical phase distortion effect

Intersymbol interference (ISI) is the impulse broadening in time resulting in interference between successive TX samples

Ref. Pictures from H. Khorramabadi presentation, Analysis and Design of VLSI Analog to Digital Analysis Analog-to-Digital Interface Integrated circuits, UC Berkeley, 2009

Linear phase filter

Nonlinear phase filter

ISI is larger in this case due to the tail in the time p response
20

The Insertion Loss Method (ILM) ( )

The insertion loss method (ILM) is a systematic way to design and synthesize a filter with a known frequency response. ILM method also allows filter th d l ll filt performance to be improved in a straightforward manner, at the expense of a higher order filter There is a historical reason why phase higher order filter. Phase information is totally information is ignored. Original filter design ignored. So the filter type must methods were developed for voice and be chosen at the very beginning human ear is insensitive to phase distortion.
21

Power Loss Ratio

Ideally PLR should be equal to 1 in the bandwidth and infinite out of bandwidth. In real filters we have to set 1PLR1+k2 over the bandwidth

22

Filter types yp
As a general rule, in the ILM, the PLR function is approximated by suitable characteristic polynomials, i.e.

PLR = 1 + k 2P 2 ( )
In particular depending on P() the most famous types of particular, filters are:
Maximally flat or Butterworth filters: Moderately linear phase response, slow cut-off, smooth attenuation in passband. l t ff th tt ti i b d Chebyshev filters: Bad phase response, rapid cut-off for similar order, contains ripple in passband. May have impedance mismatch for N even. e en Bessel filters: Good phase response, linear. Very slow cut-off. Smooth amplitude response in passband.
23

Characteristic Polynomial Functions

Low-pass case

24

Butterworth Low-pass Filters

Maximally flat in the passband: all derivatives of the transfer function till order N in =0 are equal to 0 Moderate phase distortion
Ref. Pictures from H. Khorramabadi presentation, Analysis and Design of VLSI Analog-to-Digital Interface Integrated circuits, UC Berkeley, 2009

All poles on the unitary circle with equal angles

25

Chebychev I Low-pass Filters

Equiripple in the passband Sharper transition bandwidth

Ref. Pictures from H. Khorramabadi presentation, Analysis and Design of VLSI Analog-to-Digital Interface Integrated circuits, UC Berkeley, 2009

Severe phase distortion All poles on ellipses i id th unitary circle l lli inside the it i l
26

Bessel Low-pass Filters

Flat in the passband Poor out-of-band attenuation Q asi linear phase Quasi-linear All poles outside the unitary circle
Ref. Pictures from H. Khorramabadi presentation, Analysis and Design of VLSI Analog-to-Digital Interface Integrated circuits, UC Berkeley, 2009

27

Examples of PLR for Low Pass Filters - 1 Low-Pass

28

Examples of PLR for Low-Pass Filters - 2

29

Group delay comparison

30

The Low-Pass Prototype ( yp (LPP) )

RI ZL

31

General LPP design procedure

The LPP is the building block from which real filters can be constructed. Various transformations may be used to convert it into a high-pass, bandpass or other filter of arbitrary center frequency and bandwidth. Let us consider the first circuit shown in the previous slide, and let refer to RI as the output resistance of the signal source and to ZL()=RL()+jXL() as p g ( ) ( ) j ( ) the input impedance of the filter.
2 2 (RI + RL ( ))2 + X L ( ) = 1 + (RL ( ) RI )2 + X L ( ) = 1 + k 2P 2 ( ) PLR = 4RI RL ( ) 4RI RL ( )

By equating the N coefficients of the two polynomials in shown above for a given filter (i.e. Butterworth, Chebychev or Bessel expression), the values (i e Butterworth expression) gk of either Lk or Ck for k=1,,N are determined. The values of the coefficients gk for a certain filter if both RI and RL are equal to ( t 1 (normalized values) are usually t bl d ( li d l ) ll tabled (see next slide). t lid ) In order to determine the values if RL1, RI and Lk should be multiplied times RL while Ck should be divided by RL
32

Tables examples for LPP design

Table for Butterworth LPP filters

Table for Chebychev LPP filters

33

Impedance Denormalization and Frequency Transformation of LPP - 1

As stated above, once the LPP filter is designed the impedance presented by above designed, the filter at the operating frequency can be scaled from unity to other values RL. This is called impedance denormalization. Also, either th cutoff f Al ith the t ff frequency can b t be transformed t other f f d to th frequencies th i than unity or LPP parameters Lk and Ck can be mapped to values Lk and Ck corresponding to other filter types such as highpass, bandpass and bandstop. To this purpose, we can use a new variable =f() so that:

PLR = 1 + k 2P 2 (' ) = 1 + k 2P 2 [f ( )]

Examples
1) )
' = c
c

LPF with cutoff frequency c q y LPF to HPF

' Ck =

L'k = L'k =

Ck Lk ' RL C k = c R L c RL c C k
34

2) ' =

1 c Lk RL

Impedance Denormalization and Frequency Transformation of LPP - 2

2 1 2 0 3) ' = 2 1

where

LPF to BPF

L'k =

Lk
' Ck =

Lk RL 0

Ck

' Ck' =

0 Lk RL

Ck 0 RL

' L'k =

RL 0Ck

where Note: many other transformations are possible (i.e. bandstop, notch,.)
35

Summary of the design steps

36

Example 1: Butterworth LPF p

37

Example 2: Chebychev LPF

38

Integrated planar resistors

Resistor length

Single square resistor

R=

l l l = = R[] S z w w

Material resistivity

3-square resistor
Resistor cross-section [depth(z) [d th( ) x width (w)] idth ( )]

w l
Large resistor

R[] depends on the manufacturing process and on the material employed (between 0 025 0.025 /square up to 100 /square) l/w l/ can be controlled by the IC d i b t ll d b th designer with ith high accuracy Large resistors are expensive in terms of area

39

Integrated planar capacitors

Capacitor area [length (l) x width (w)]

C =

S = (w l ) t t
Dielectric thickness

Dielectric constant (usually SiO2)

/t depends on the IC manufacturing process wl can be controlled by the IC designer with high accuracy y g g y Large capacitors are expensive in terms of area. Area reduction can g q (e.g. g be achieved using various techniques ( g through trench or stacked capacitors)
40

Lateral flux capacitors p

Lateral flux capacitors are designed to exploit lateral electric fields. The idea is to increase the capacitance at no price in terms of planar area. Fractals have the additional advantage of low series resistance and inductance higher Q, better robustness to the etching process.
41

Integrated monolithic inductors

The structure of a monolithic inductor consists of a spiral of different shape (rectangular, circular, hexagonal or octagonal) No clear analytical formulas exist to estimate L In the case of L. square inductors L is approximately given by:
4.7 nH

L 1.3 10 7
Total area of the inductor

Am S
1/ 6

5/3

Total area of the metal track

1/ 4

7/ 4

(G + w )

33 nH Width of the metal track

Distance between coils

Performance of a monolithic inductor are limited by 3 main parasitic effects: Metal wire resistance (reduced using circular shape and by removing the most internal 4 or 5 turns of the coil) C Capacitive coupling ( d iti li (reduced b i d by increasing th di t i the distance between inductor and chip surface) Magnetic coupling (reduced by using a p g p g( y g patterned g ground shield (PGS) between the inductor coil and the chip surface)
42

Quality factor of integrated inductors

Q is measured at frequencies of operation (typically > 1GHz) (y y ) Hard to have Q larger than 10 with CMOS technologies Other substrates or techniques are required for better performance
43

Monolithic Transformers 1
Monolithic transformers have been used extensively in RF circuits. They usually consist of 2 inductors combined in different way. For different transformer structures, the coupling coefficient k, the turn ratio n, and other parameters may vary considerably. Depending on whether the lateral or vertical magnetic coupling i used, t ti li is d transformers can f be planar or stacked. Planar can be symmetric or asymmetric asymmetric. Usually n is small and coupling k up to 0.7.

44

Monolithic Transformers 2

In stacked transformers, both vertical and lateral magnetic coupling is used Pros: high self-inductance and coupling coefficient (k=0 9) smaller size (k=0.9), Cons: high-parasitic capacitance, poorer quality factor

45

SAW filter basics - 1

Surface Acoustic Wave filters are based on interdigital trasducers (IDT) L Largely used f RF and IF fil l d for d filters b but not on the same chip Piezolectric substrate quartz, tourmaline, gallium phosphate, lithium niobate, lithium tantalate and piezoceramics Etched Aluminium pattern on top RF electric energy is transferred to the crystal and turns into vibrational Vibration propagate along the surface
Longitudinal waves: ~6000 m/s Transversal waves: ~3000 m/s

Ref. Pictures from D. Morgan, Surface Acoustic Wave Filters with applications to electronic communications and signal processing, Elsevier, 2007

46

SAW filter basics - 2

Piezoelectric crystals have some ideal properties for wave propagation:
Anisotropy Low losses (if we have a pure crystal) Ability to propagate signals at y p p g g frequencies larger than 1 GHz

Best coupling achieved when the electrode width is in the order of /4 By properly designing the pattern of the output IDT only some frequencies are turned back into RF signals bandpass behavior Every stage may cause reflection and a delay FIR filter design techniques Linear phase response
Ref. Pictures from D. Morgan, Surface Acoustic Wave Filters with applications to electronic communications and signal processing, Elsevier, 2007

47