ISSCC 2005 / SESSION 17 / RF CELLULAR ICs / 17.8
17.8
A 5th-Order Continuous-Time HarmonicRejection GmC Filter with In-Situ
Calibration for Use in Transmitter
Applications
Jacques C. Rudell, Ozan E. Erdogan, Dennis G. Yee, Roger Brockenbrough,
Cormac S. G. Conroy and Beomsup Kim
17.8.3, to find the frequency at which the phase through the loop
equals 360o. For this configuration, the loop gain when the phase
reaches 360o is much greater than 1, thus satisfying the condition
for oscillation. The oscillation frequency ωosc is given in Fig. 17.8.3
and is dependent exclusively on values of Gm and C in the filter.
Therefore, an estimate of component values over process and
temperature can be obtained given that a ratio between the oscillation frequency and the filter –3dB frequency exists; this ratio is
approximately 1.5.
Berkäna Wireless, Campbell, CA
The advent of sub-micron CMOS processes has ushered in a new
generation of highly integrated, low-cost communication systems
for a variety of applications from single-chip wireline Gigabit
Ethernet transceivers to monolithic wireless front-ends. Reduced
form factor and lower cost are two of the well known benefits of
integration in CMOS. However, additional benefits are derived
through the integration of a mixed-signal system in CMOS, by
allowing new digital calibration techniques to enhance the analog
circuit and system performance. As an example of these concepts,
this paper presents a continuous-time filter with a new, in-situ
filter calibration method. The intended application of this filter is
to suppress all higher-order harmonics generated by both a modulator and feedback mixer within a GSM offset-PLL (OPLL)
transmitter [1]. This paper describes the OPLL filtering requirements, the implementation of the filter, followed by a description
of the filter tuning technique along with measured results.
A key block of any OPLL is the harmonic-rejection filters which
precede the PFD shown in Fig. 17.8.1. Specifically in the GSM
standard, the modulation mask is greatly influenced by these filters, particularly the requirement of –60dBc, 400kHz offset from
the carrier. The filter attenuates unwanted energy produced at
all the higher-order harmonic frequencies generated by the mixing of LOs in the OPLL reference and feedback paths. These harmonics, if left unfiltered, would pass through the hard non-linearity associated with the PFD, intermodulate, and create undesired energy in the wanted signal band. This is illustrated with
the 3rd and 5th harmonics in Fig. 17.8.1.
Based on system simulations, the filter is implemented as a cascade of one real pole followed by two biquad stages where all
poles are realized with a differential Gm-C approach, as shown in
Fig. 17.8.2. The filter must be tunable from 60 to 150MHz with
5% tuning accuracy. Each capacitor is tunable with a fixed base
capacitance of 250fF in parallel with a 6b capacitor array, binary
weighted with an LSB of 25fF. Coarse filter band selection is realized by varying the degeneration resistance through a 3b resistor
array in all Gm stages, Fig. 17.8.2.
All the capacitors are equal in size with the same capacitor settings in each array. Therefore, the frequency response is defined
by the ratio between Gm stages within the biquad. To ensure good
matching between Gm cells and an accurate biquad Q over temperature and process, a unit Gm cell is created. A different
transconductance from the unit cell is derived by scaling all
devices by a factor of α, where α corresponds to the ratio between
the desired value of Gm to the unit Gm, Fig. 17.8.2.
This paper provides a salient approach to filter tuning applicable
to a broad class of filter applications. In this tuning approach,
prior to normal TX operation, the filter biquads are reconfigured,
creating positive feedback and transforming the filter into an
oscillator, as shown in Fig. 17.8.3. The output of this oscillator is
then digitized with an inverter and a set of counters are used to
estimate the oscillation frequency relative to an accurate crystal
reference. A closed-form solution may be obtained for the oscillation frequency of the filter by evaluating the configuration in Fig.
322
Prior to normal TX operation, this filter is calibrated. Switch
SW1 is opened while switches SW2 are closed. The oscillation frequency of the filter is then measured by counting the number of
rising edges produced by the oscillator during a fixed period, as
determined by another counter clocking off a crystal reference
source. With an estimate of the oscillation frequency, all the
capacitors in the filter are adjusted. The process of measuring the
oscillation frequency and tuning the capacitors is then repeated
through a method of successive approximation until the targeted
oscillation frequency, and ultimately the filter corner frequency,
for a targeted IF is achieved. SW1 and SW2 are then set for normal filter operation.
This tuning approach differs from previous master-slave methods
[2] which require extra hardware and rely on matching between
the tuning element and the filter components. Other attempts at
in-situ filter tuning require additional hardware to generate calibration tones injected into the filter [3]. Two primary advantages
exist for this filter tuning method. First, the filter is being tuned
in situ. Stated differently, the devices being tuned are the same
elements which will be used by the filter during normal operation. Second, a relatively small amount of digital hardware is
required to tune this filter, thus minimizing the calibration hardware and chip area.
Figure 17.8.4 shows the measured filter frequency response, the
filter tuning range, BW resolution of 2%, and demonstrates the
filter capability of retuning over temperature, and returning to
the targeted BW. In Fig. 17.8.5, the TX output spectrum is shown
when the filter is in the narrowest BW setting, widest BW, and
the ideal target BW after calibration. Note that in the widest BW
setting, the modulation mask is degraded due to intermodulation.
The modulation mask is met with margin in the narrowest BW
state, but the wideband OPLL noise is higher than the attenuation of the desired signal if the filter BW is too narrow. Finally,
Fig. 17.8.6 illustrates the tradeoff between the modulation mask
performance at 400kHz offset and the inband integrated OPLL
phase noise as a function of the filter capacitor tuning codes,
where a low number on the x-axis corresponds to the widest filter
BW. The post calibration filter setting is marked and represents
the optimal setting for the overall TX performance. This filter is
integrated with the entire transmitter in a 0.18µm CMOS technology. Die micrograph with a summary of data is shown in Fig.
17.8.7.
Acknowledgements:
The authors thank Julian Tham, Ham Lenh and Jin-Su Ko for their contributions.
References:
[1] T. Yamawaki et al., “A 2.7-V GSM RF Transceiver IC,” IEEE J. SolidState Circuits, vol. 32, pp. 2089-2096, Dec., 1997.
[2] M. Banu and Y. Tsividis “On-Chip Automatic Tuning for a CMOS
Continuous-Time Filter,” ISSCC Dig. Tech. Papers, pp. 286-287, Feb.,
1985.
[3] H. Khorramadabi et al., “Baseband Filters for IS-95 CDMA Receiver
Applications Featuring Digital Automatic Frequency Tuning,” ISSCC Dig.
Tech. Papers, pp. 172-173. Feb., 1996.
• 2005 IEEE International Solid-State Circuits Conference
0-7803-8904-2/05/$20.00 ©2005 IEEE
Authorized licensed use limited to: University of Washington Libraries. Downloaded on December 3, 2009 at 17:03 from IEEE Xplore. Restrictions apply.
ISSCC 2005 / February 8, 2005 / Salon 9 / 4:30 PM
Single Differential Biquad used in both stages
Offset Phase-Locked Loop
Modulator
I BB
HR Filter
fIF
Q BB
Loop
Filter
fIF
VCO
PFD
C1
Reference Path
fIF
External
Power
Amp
HR Filter
I LO
Q LO
Feedback Path
C2
Local
Oscillator
α
2nd Order
BiQuad
α
α
+∆Vin
IF +f
m
To Common
Mode Feedback
α (W/LMN5)
f
(W/LMP1)
α
(W/LMN3)
Rdegen
α
Figure 17.8.1: OPLL with HR filters and the effect of problematic harmonics.
3-bit Band Selection Network
Band Select<0>
+∆iout
(W/LMN4)
Band Select<1>
-∆Vin
Band Select<2>
To CMFB
(W/LMN6)
Figure 17.8.2: Single biquad stage and scalable Gm cells with
band-selecting Rs.
Filter Range
Filter Resolution
Two Cascaded Biquad Stages
SW1
- gm11
gm41
Filter
In
gm21
C1
SW2
- gm12
- gm31
gm22
gm42
C2
C3
Filter Cal.
Out
SW2
α ⋅ (2ID /VDsat )
=
= α ⋅Gm(unit)
) 1 + (2ID /VDsat ) ⋅ Rdegen
(W/LMN2)
RDegen / α RDegen / α
α
α ⋅ (2ID /VDsat )
1 +α ⋅ (2ID /VDsat ) ⋅ (
(W/LMP2)
α
α
(W/LMN1)
C4
5 *f
3 *f
fIF -3
IF -f
m
*fm
fIF +f
m
2nd Order
BiQuad
er
1 Real
Pole
-∆iout
Intermodulation of 3rd
and 5th impacting the
modulation mask
HR Filt
|Sx(ω)|
Gm =
Scalable Gm cell
Unfiltered Single-Tone
Spectrum @ PFD Input
3 Stage Gm- C Filter
C = C1= C2= C3 = C4
C3
- gm32
Filter
Out
1 LSB Step.
3MHz (2%) Tuning
Resolution
C4
Ȧosc =
48MHz – 184MHz:
100MHz Tuning
Range
gm22 ⋅ gm12 ⋅ gm31
(gm32 + gm31 ) ⋅ C 2
Hot Temp. Cal. (100o C)
Cold Temp. Cal. (-30o C)
All Digital Calibration Engine
17
RAM Look Up
11 bit Counter
Target Count
7 bit Counter
CLK
REF_CNT<7:0>
Successive
Approximation
Calibration
Engine
Figure 17.8.3: Filter mode (SW1 closed, SW2 is open), oscillation mode
(SW1 open, SW2 closed).
Att
25 dB
SWT 10 ms
0
c2
IFOVL
1 SA
VIEW
2 SA
VIEW
3 SA
VIEW
LIMIT CHECK
PASS
-10
Delta 1 [T1 ]
-67.11 dB
400.000000000 kHz
Marker 1 [T1 ]
-1.61 dBm
848.800000000 MHz
400kHz Mask Performance vs. In-Loop BW PLL Noise
-56
A
TX Modulation mask requirement
from the GSM Standard
-20
-30
-40
Too wide a filter BW
allowing intermodulation
-50
SWP
200 of
200
-60
Too narrow a filter BW
which attenuates the
carrier, raising the
PLL noise floor
1
-70
1.20E-04
400kHz Modulation (dBc)
GSM Modulation Mask Spec @ 400kHz
TX Modulation mask
requirement from
the GSM Standard
-58
-60
NoisePower 1MHz from Carrier
1.00E-04
8.00E-05
Post-Filter Calibration
Code of 36 for this
TX carrier
-62
6.00E-05
-64
4.00E-05
-66
2.00E-05
-68
-80
-90
Figure 17.8.4: Filter freq. response, range, resolution, hot & cold post cal.
(dBc)
* VBW 30 kHz
0 dBm
Post Calibration
Room Temp.
Post Calibration
Room Temp.
Mod. Mask @ 400kHz offset RBW=
* RBW 30 kHz
Ref
1
Post Calibration
Hot Temp. (100o)
Post Calibration
Cold Temp. (-30o)
0.00E+00
0
Ideal filter setting
after calibration
ResBand=30kHz
OSC_CNT<10:0>
XTAL Osc.
Noise Power (mWatts) @ 1MHz
CLK
10
20
30
40
50
60
70
Filter Capacitor Code Value
-100
Start
848.8 MHz
180 kHz/
Stop
850.6 MHz
Figure 17.8.5: Impact of the filter on the GSM TX modulation mask.
Figure 17.8.6: 400kHz modulation mask versus inband OPLL noise.
Continued on Page 601
DIGEST OF TECHNICAL PAPERS •
Authorized licensed use limited to: University of Washington Libraries. Downloaded on December 3, 2009 at 17:03 from IEEE Xplore. Restrictions apply.
323
ISSCC 2005 PAPER CONTINUATIONS
Reference Feedback
Path
Path
HR Filter HR Filter
Filter Performance
Minimum Filter BW
48MHz
Maximum Filter BW
184MHz
Filter resolution Max. BW
3.1MHz (2 %) @ 151MHz
Filter resolution Min. BW
701kHz (1.5%) @ 48MHz
IIP3
7dBV
Filter Input Noise inferred from TX measurement
9.3µVRMS (30kHz BW)
Worse Case Calibration Time
90µs
Measured Rejection at 3rd and 5th Harmonic
45dB (or greater)
Maximum Output Operation Voltage Swing
300mV (differential)
Current from a 1.8Volt supply
7mA (12.6mW)
400kHz Modulation Mask Performance across all bands
-65dBc (Worst Case)
Figure 17.8.7: Filter die micrograph and data.
601
• 2005 IEEE International Solid-State Circuits Conference
0-7803-8904-2/05/$20.00 ©2005 IEEE.
Authorized licensed use limited to: University of Washington Libraries. Downloaded on December 3, 2009 at 17:03 from IEEE Xplore. Restrictions apply.