• 沒有找到結果。

A frequency estimation algorithm for ADPLL designs with two-cycle lock-in time

N/A
N/A
Protected

Academic year: 2021

Share "A frequency estimation algorithm for ADPLL designs with two-cycle lock-in time"

Copied!
4
0
0

加載中.... (立即查看全文)

全文

(1)

A

FREQUENCY ESTIMATION ALGORITHM FOR

ADPLL

DESIGNS

WITH

TWO-CYCLE LOCK-IN

TIME

Chia-Tsun Wu, Wei Wang, I-Chyn Wey, An-Yeu (Andy) Wu

Graduate Institute of Electronics Engineering, and Department of ElectricalEngineering, National Taiwan University, Taipei 106, Taiwan, R.O.C.

Abstract-This paper presents a Frequency-Estimation This paper presents a Frequency-Estimation Algorithm Algorithm for the ADPLL designs instead of traditional (FEA) for frequency-acquisition in two-cycle lock-in time, binary frequency-search algorithm. With the proposed with stable output frequency. We proposed a close-form ADPLL architecture and synchronization process, the lock analysis in time domain than others approach. A new VLSI time can be optimized to two cycles. As the reference clock architecture of ADPLL is also developed and realized in varies or frequency multiplication switches, lock time holds UMC 0.18

lp6m

CMOS process. Cooperated with

intwo reference clock cycles. An implementation of proposed proposed scalable digital lock detector, the proposed ADPLL design is realized in UMC 0.18 pm 1P6M CMOS ADPLL is suitable to fit demands of different design technology with core area of 520x530 pm2. The PLL has the applications.

frequency range of 140 MHz to 1030 MHz with 22ps DCO resolution.

Reference Clock

1. INTRODUCTION frePFD LoopFilter

PLL (Phase-Locked Loop) is widely used for various

applications, especially in wire/wireless communications,

FrequencF

o

DigitallyOControlled)

CDR (Clock Data Recovery), and clock generators. In

FaOutput

Clocllfo

recent years, SoC (System-on-Chip) becomes a major Figure 1 AtypicalADPLLblockdiagramwith

trend. The ADPLL (All-Digital Phase-Locked

Loop)

r.

frequency-applies digital design methodology to avoid mixed-signal design problems, and is suitable to be packed as hard IP,

firm IP, and soft IP. These advantages make ADPLL very II.PROPOSEDFREQUENCY-ESTIMATIONALGORITHM attractive to beintegratedinto SoCdesigns. The main feature ofFEA is based on

1)

Correlations

Another ADPLL advantage is that the ADPLL usually between reference clock and DCO characteristics; 2) employs frequency-searching algorithm to achieve high- Analysis of DCO control code in time domain. The desired performance frequency lock-in time [1][2][4] as shown in frequency output

f,

canbe obtained by further calculation Fig. 1. This fast-locking ADPLL is very useful for instead of traditional search.

frequency hopping based FHSS (Frequency-Hopping In this context,

frf

and

f

represent the

frequencies

of Spread Spectrum) applications such as 802.11, BluetoothInticoex,read epsnthefqucesf

SpeandFHMAe (FrequenaicyHoping Multiple802.1

,

Access)to%

reference

clock,

and desired

output

frequency,

respectively.

UWBFHMA

(UltraeWideBand)

ince

Multh

ple

Ac

inss) af

N is

multiplication

factor. The

frequency

multiplication

UW

(lta

idetransceiver i

function

ofa PLLinlockedstateis

described

as

FHSS/FHMA system switches carrier frequencyrapidly, a fast-switching and stable frequency synthesizerto generate

thehopping carrier frequency is essential to FHSS/FHMA fo Nxfre1 (1)

transmitters and receivers.

Parameters aredefined for clarification:

MostanalogPLLs provide higher frequency resolution,

-ref

The cycle time of reference clock.

but suffer from slow acquisition time. Lots of ADPLLs

Tmi,

The minimum

cycle

time of DCO when DCO employ Binary Search Algorithm (BSA) as frequency- operatesatthe maximum

oscillating frequency

search algorithm torour 1 * *achieve faster lock-in time. The BSA1 1 r 11 *1 rT^1 1 R

RmansThe maximum

The maximumcycle

time ratio

oscillatin fequen cy

between reference

[2][4]

is a very simple, and useful algorithm. The search clock and DCO clock.

(DCO

operated

at

Tmi,)

complexityis 0(log2n). Using* r 1 1 1the BSArT^1 1 1 * *canreduce lock-in T;~maxc The maximum cycle time of DCO when DCOclc

andmDC

clo

(DCO

opaD

at DCO

time to 16 reference clock

cycles [3].

The lock-in time can operatesatthe

minimum

oscillating

frequency. be

improved

more than 40 times faster than traditional R Tc

analog PLL [5]. Another approach for frequency-search

uses TDC (Time-to-Digital Conversion), and developed clock and DCO clock. (DCO operated at 'max) DCO~~~~~~~arhtetr [1oaheesvnccelc-ntm. Then, the correlations between reference clock and DCO

lock-in

range are deducedas

(2)

Rmin Tmax =Rma Tmin =ref (2) feature of proposed FEA is, even under temperature

Rm*n'ma

-RaxmJ

rref2 fluctuations, power supply or process variations, the FEA

Transformations from (2) givesexactlythe desired control code.

Assume an ADPLL operatesfrom 100MHz to 500MHz

_Tref

- ref with 14 bits binary control code of DCO. The behavior

Tmax

Rmm

TmiM-

Rmax

simulations of proposed algorithm and BSA are shown (3) within Fig. 2. When the BSA is applied in the frequency Consider a DCO provides Tfrequency acquisition steps. acquisition, the step size will be divided by two after for Then, we can derive the relation between resolution and each frequency-search. Output frequency is unstable during difference of and

zmin

as convergence process in Fig. 2(a). The proposed FEA locks the reference clock in very few

cycles

with a stable

Tref

~ convergence process asshown in

Fig.

2(b).

Tref ref

Ar - Tmax

-min

Rmin

Rmax

400

T T

(4)

350

J

B00

~300 '0

whereJXis the DCOacquisitionresolution, and Tis the

total number of DCOacquisitionsteps. 100

200

After ADPLL closedloop converges, the cycle time of * 10o 0 0

~~~~~~~~~~~0 10 fr °f30B s40 50 ( a 0 10 t?OefC1SlS 40 50(b

outputfrequencycanbedescribedas eV

(b)

1 Figure 2. (a)Behavior simulation ofbinary-searchalgorithm. (b)

=FxAr+ m Improved convergence of

frequency-estimation

algorithm.

(5)

whereFis the desired controlcode foraccurate

frequency.

III. PROPOSED ADPLL ARCHITECTUREFOR

TWO-CYCLE

LOCKING

From(1),and(5) The VLSI architecture, Fast-Lock Engine (FLE), is

developed to realize FEA algorithm in Fig. 3(a). With Nx

(F

xA T+Tmin)

Tref

(6) cooperation of FLE, an ADPLL design can provide

fast-lockoperation. Equation(6)denotes therelationshipbetween reference

clock and DCO parameters after closedloopconverges.

F--

--L-ck

Engine

From

(3), (4),

and

(6)

wehave (a) InnerDCOHoperatedatFa Inner DCOLoperatedatF.

Iefreref

C Cycle CounterR, CycleCounterRmin

ref ref.1

Lock Fast

Frequency-kXF

Rin Rmax lefr = T-J O

Nx!FxR 7 +; J0rf 7 ()

'rref

Detector Estimation

~~~~~~~~~~~~~~System

& DCO

Controller

PFD LoopFilter

Afterreorganization,the desired DCO control codeFis

expressedas FeedbackClock FrequencyDivider Output Cock,fo OutputDCO

(T x

Rmax

X

Rmin)

-(Tx

Rmin

xN)

TyjicADRL

Lo

F-

NX(R

-R)

Figure3. ProposedADPLLarchitecture fortwocycles lock.

NX max

-kin)J(8

When targeting at frequency acquisitionwith two-cycle

Note that theTmax, Tmin, andJS areparameters of process

lock-n

time,inx

an

rmun

shou

be

concu

e

dependence,

and difficult to be identified before

avai

Hene,

two

addioain

DCOs are uilized in fabrication. Also, the exact value of

i+ef

varies between

proposed

ADPLL

configtion.

inner

DCOH

and

u

dCO

applictions

Bu.m

n mncnb bandb

proposed

ADPLL

configuration.

Inner DCOH and DCOL

applicaions.

But Rmax and Rmin can be obtained by were operated at the highest / slowest frequency

frequency counters and dividers during run time. Also, the g q

y,'

TiSone ofpre-defined parameters of theDCO, and NiSthe repcilyThgnradcoksgasaetasmtd mutpicto fatr Bae oncreae

'eedny

separately to cycle-counters.

While

system controller * . '

~~~~~~~~receives

Rmax, and Rmin from two cycle-counters, the F can can b calulatdbypropsed FA (8. Animpotant be calculated for

desired

frequency

according

to FEA.

(3)

A. Fast-Lock Process

fd1vF

ll~ QO QI locked

Fig. 4(a)

illustrates

synchronization

between reference lD D DD c

clock,

andfo.

Assumea

frequency

step is

applied

tothe

fref

-nQO DelayCell oQI CycleSync

at cycle

1

inFig. 4(b). The LD(Lock Detector)detects the > (Unlocked)

frequency step during cycle I and activates trigger-signal frd II

(CycleSync). The Rmax and Rmin are both calculated

(a)

simultaneously. TheDFFsuspends DCO on therising edge

ofcycle 2 andFLE is now activated. FLE calculates theF

finput

waveorm truthtable

during cycle

2. At

last,

the DCO resumeswithnew control fref JLF Qi QO phaeerror staus

codeFattherising edge of cycle 3. Based on this process, 22r-1 0 0 <

1/2Trejt

TLD unlocked fast-lock frequency acquisition can be realized in two f 0 1

<cTLD

d(ed

cycles.

div

-- 1 0

Y/2rei

TLD unlocked

Fast-lock Normal _ 1 1 >Y/2reI TLD unlocked

ControlCode,F Control Code

T D (b)

fref |,dk rL T Digitally-Controlled

ocked

rdear Oscillator

-CycleSync ( .ha.,[

.phaseerror

- Cycle 1I - Cyde2 - - Cyde3 -LD3 -TLD2 -TLDI TLDI TLD2 TLD3

CYdeSJc (c)

to

Figure

5.

(a)

Lock detector

architecture.

(b)

Charateristics of

LD

(b)

(c)

Simulation of LDoutputv.s.different threshold.

Figure4. (a) Synchronizationfor DCO.(b)Fast-lock Process. C.

Workingflow

chart

oftheproposedADPLL

Fig. 6 shows theproposedADPLLworking flow chart. B. ConfigurableLockDetector (LD) Once the phaseerror over the threshold, LD activates FLE,

AnewLDconfiguration is utilized asphase /frequency the system controller receives the desired control code F.

lock detector as shown inFig. 5(a).Phase error

betweenfief

The fast-lock

acquisition

is

completed

in two

cycles.

andfdivand circuit status are shown in Fig. 5(b).The main Otherwise, fine phase acquisition and are performed by function ofLD is the detection ofnumerous phase errors. phase detector

(PD); phase

/

frequency

maintenance are Once the phase error exceeds the threshold TLD, the LD controlled

by

ADPLLclosed

loop.

triggers CycleSync and activates the FLE for fast-lock

operation. Thedelaycellcanbe constructed withascalable Lockin

Sats

buffer chain. Thus, the propagation delay of delay cell decides threshold ofLD. The relationcanbeexpressedas

TLD =MXTBUF (8)

1 (8) Doutput~~UDn F-cstLockCyde1

where TLD is the LD threshold, Mis the number of buffer I

CdculateR,,,

R,in stage,andTBuFisthegate delayofabuffer. LoopFiIter LoopFitter LaCCydue2e

Also, the total umber of buffer stage,

M,

should be W DibleDCOOsdllainq

determined by application specifications. The relation of Accderae Decederate Enabe

LD outputversus different thresholds is illustrated inFig. DCO DCO DCOOsciIlating 5(c). For a popular application example of video, which

referencingaHSYNC(horizontalsynchronization)

asfief

to

generate 27MHz clock, 5 nsphase errormaybe tolerated. Figure6. ADPLLworkingflowchart.

When referencing a HSYNC from VGA to generate 200 MHz clock for LCD monitor, the phase error may only be

tolerated less than 500 ps and M should be reduced. The IV. IMPLEMENTATION AND SIMULATION

proposed LD is scalable and flexible to meet demands of Based on the proposed algorithm and architecture, a

various

applications,

prototype ADPLL design is designed and realized using

(4)

UMC 0.18pm 1P6M CMOS technology, as shown in Fig. 3 250

7. For linear acquisition step is desirable in proposed 2

200

frequency-estimation algorithm, a high-linearity DCO l

design [6]

can be cited for purpose. With

post-layout

0

simulations, the contains frequency acquisition

-10

l0

0

steps

and resolution is about 22 ps. The DCO

operates

from 2 50

140MHzto 1030 MHz. Three DCOsare

replicas

forFLE. -3 01__I

The ADPLL core area is

520x530

'm . 2 efr

fCyP

8 1 (a) 20 fref Cycles (b)

Figure

9.

(a)

Phaseerrorv.s.reference

clcok cycles.

(b)ADPLL

frequency out v.s reference clock cycles.

_...

V.CONCLUSIONS

DCOH The Frequency-Estimation Algorithm for ADPLL and

m= [DC0L . VLSI architecture

are

presented

inthis paper.

Compared

to

existing algorithms, proposed algorithm

realizes the fastest lock-in time of two

cycles. Also,

a new scalable LD is utilized for

requirements

of different

applications.

The prototype

design

is

implemented

in UMC 0.18 CMOS I

~

I

I~

I I I I I I I I I I I IIprocess with standard-cell

library.

The

chip

operates from 140 to 1030MHz, and the core area is 520x530

ptm2.

The fast-lock ADPLL design can be easily transferred and incorporated into system integration due to its digital

Figure

7.

Chip layout

of theprototypeADPLL.

(where

DCO:

feature.

outerDCO, DCOH: inner DCOoperatedatF, DCOL:inner DCOoperatedatFmin,)

Acknowledgment

For example of fast-lock operation is shown in Fig. 8. The authors thank CIC (Chip Implementation Center)

Inthe initial state, reference clock is 15.625 MHz, N is 32, for CAD tool supports and technical discussions. output clock frequency is 500 MHz, and corresponding

control codeFis 50. AsmultiplicationnumberNswitches

to 48; the FLE is activated; desired F should be 16. The References

output frequency turns out to be 750 MHz in two cycles [1] Takamoto Watanabe andShigenoriYamauchi, "AnAll-DigitalPLL

and the acquisition process is exactly verified with for Frequency Multiplication by4 to 1024 With Seven-CycleLock

simulation inFig. 8. Time,"

2003.

IEEE Journa of Solid-State

Circuit,

vol.

38,

no.

2,

February

For a practical industrial application, Fig. 9(a) shows [2] C. Chung and C. Lee, "An all-digital phase-locked loop for

high-the phase difference versus reference clock for 200 MHz.. ~~~~~andSystems,speed clock generation,"vol.3, pp.26-29,IEEE International Symposium onMay 2002. Circuits

frequency

output. The

phase

error at the

rising edge

of an Sytes vo.3 p2-9 a 02

frlequenc

3isoutput.0The phasea

error

atg.

taThe

risin e o [3] I. Hwang, S Lee, S. Lee, and S. Kim, "A digitally controlled

phase-cycle 3 iS around 0.047 radian in Fig. 9(a). The desired locked loop with fast locking scheme for clock synthesis

outputfrequencyis shown inFig. 9(b). application," IEEE Int. Solid-State Circuits Conf. Dig. Tech.

Papers,Feb.2000,pp. 168-169

FZF [4] T.-Y. Hsu, C.-C. Wang, andC.-Y. Lee, "Design and analysis of R[9: 24 portablehigh-speedclockgenerator,"IEEE Trans.CircuitsSyst. II,

F19:0] vlo.48, pp. 367-375,Apr.2001.

CLg|_ [5] H.-TAhnand D.J.Allstot,"Alow-jitter1.9V CMOS PLLfor

ultra-CLjjj SPARCmicroprocessor applications,"IEEE K. Solid-StateCircuits,

koutliuI

vol.

35,

pp.

450-454,

May

1999.

[6] Chia-Tsun Wu, Wei Wang,I-Chyn Wey, andAn-Yeu(Andy)Wu,

"A Scalable DCO Design for Portable ADPLL Designs," IEEE

Figure8. Simulation offrequency multiplicationswitch. InternationalSymposiumonCircuits and Systems,pp. 5449-5452,

May2005.

數據

Figure 4. (a) Synchronization for DCO. (b) Fast-lock Process. C. Workingflow chart oftheproposedADPLL
Figure 8. Simulation of frequency multiplication switch. International Symposium on Circuits and Systems, pp

參考文獻

相關文件

We solve the three-in-a-tree problem on

Here, a deterministic linear time and linear space algorithm is presented for the undirected single source shortest paths problem with positive integer weights.. The algorithm

In particular, we present a linear-time algorithm for the k-tuple total domination problem for graphs in which each block is a clique, a cycle or a complete bipartite graph,

A spiral curriculum whose structure allows for topics or skills to be revisited and repeated, each time in more detail or depth as the learner gains in knowledge, skills,

– The futures price at time 0 is (p. 275), the expected value of S at time ∆t in a risk-neutral economy is..

Compensates for deep fades via diversity techniques over time, frequency and space. (Glass is

* All rights reserved, Tei-Wei Kuo, National Taiwan University, 2005..

synchronized: binds operations altogether (with respect to a lock) synchronized method: the lock is the class (for static method) or the object (for non-static method). usually used