This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE JOURNAL OF SOLID-STATE CIRCUITS 1

A Dynamically Biased Multiband 2G/3G/4G Cellular

Transmitter in 28 nm CMOS
Siddharth Seth, Dae Hyun Kwon, Sriramkumar Venugopalan, Sang Won Son, Member, IEEE, Yongrong Zuo,
Venumadhav Bhagavatula, Member, IEEE, Jaehyun Lim, Dongjin Oh, and Thomas Byunghak Cho, Senior
Member, IEEE

Abstract—We present a highly configurable, low-power, peak-to-average-power-ratio (PAPR) signals in the case of 4G
low-area, low-EVM, SAW-less transmitter (TX) architecture that LTE where the use of OFDM and complex modulation schemes
is based on a dynamically biased power mixer. All FDD/TDD pushes the PAPR to >6 dB. Within 4G LTE20, the baseband
bands from 0.7 to 2.7 GHz for 4G LTE Rel-11 and 3G HSPA+
are supported in addition to 2G quad bands. The power-mixer signal can be either wideband where the 18-MHz channel is
bias current is dynamically adjusted based on the instanta- fully occupied by 100 resource blocks (RB) of 180 kHz band-
neous baseband signal swing using a fully-differential hybrid width (100 RB/full-RB), or narrowband where all the signal
full-wave rectifier/envelope-detector circuit. Dynamic biasing power is concentrated in a single RB (1 RB). When a sin-
leads to greater than 50% current savings when compared to gle or few RBs are transmitted close to the channel edge, the
fixed-biasing while providing a higher output power with better
linearity. Implemented in 28 nm CMOS technology, the TX TX nonlinearity leads to the generation of third- and higher
shows better than −157 dBc/Hz RX-band noise emission and order counter-intermodulation products (CIM3, CIM5, etc.) in
−41 dBc ACLR for output powers up-to +4 dBm across all the adjacent bands, causing the TX to fail spurious emission
3G/4G bands, while demonstrating above 80 dB of gain control specifications. In addition, the TX noise performance is criti-
range. In addition, the TX can be configured to provide better cal, especially for FDD systems where the receiver (RX) and
than −65 dBc CIM3, allowing it to meet stringent spurious
emission specifications when transmitting 1 RB 4G LTE signals in the TX are turned ON and operating at the same time. The out-
B13/B26/B1. of-band noise from the TX can fall into the RX band and reduce
Index Terms— ACLR, active mixer, baseband, biquad, CIM3, the RX sensitivity, effectively worsening the RX noise figure.
CMOS integrated circuit, dynamic biasing, E-UTRA, EVM, high The two standard architectures used in cellular TX design are
efficiency, high-speed packet access (HSPA), long-term evolution the voltage-mode passive-mixer-based driver amplifier (DA)
(LTE), low power, mobile communications, out-of-band noise, architecture [1]–[4] and the current-mode active-mixer (also
peak-to-average-power ratio (PAPR), radio frequency, transmit- known as power mixer) based architecture [5]–[8]. The passive-
ter, 2G, 3G, 4G.
mixer-based upconverter, while being more power-efficient, is
I. I NTRODUCTION less linear [5]. In the passive-mixer and DA architecture, the
DA worsens the CIM terms generated by the passive mixer.
I NTEGRATED multimode multiband transmitters (TX)
need to meet diverse specifications related to output power,
spectral regrowth, spurious emissions, out-of-band noise, and
CIM3 can be improved by removing the DA by utilizing a
current-mode power mixer as discussed in [5]. However, such
gain range, while occupying low area and maintaining high a traditional, power-mixer-based TX is biased in Class-A mode
power-efficiency. The push to support multimode multiband (fixed bias). In such fixed-bias systems, the signal-path blocks
power-amplifier modules in today’s mobile handsets has neces- are biased with fixed currents that are high enough to allow the
sitated an increase in the required RF output power from the peak signal swings to pass through with good fidelity. As the
TX. This increases the dc power consumption of the TX RFIC PAPR goes up, these peak swings happen more infrequently,
and improving the power-efficiency of the TX path becomes further reducing the power efficiency of the fixed-bias systems.
critical for improving the mobile handset’s battery life. In this paper, we propose a dynamically biased TX archi-
Multimode transmitters are required to process constant tecture [9] in which the bias current is automatically adjusted
envelope signals in the case of 2G GMSK, as well as high based on the demands imposed by the baseband signal swing,
leading to a very power-efficient design. Dynamic biasing
Manuscript received September 07, 2015; revised October 20, 2015 and
reduces the average bias current in the power-mixer stage,
December 07, 2015; accepted December 10, 2015. This paper was approved
by Guest Editor Salvatore Levantino. which allows reducing noise and improving linearity. The
S. Seth, S. Venugopalan, S. W. Son, Y. Zuo, and V. Bhagavatula paper is organized as follows. In Section II, we introduce
are with Samsung Semiconductor, San Jose, CA 95008, USA (e-mail: the concept of dynamic biasing in a power-mixer architecture
[email protected]; [email protected]; sangwon.s@samsung.
com; [email protected]; [email protected]).
and compare it to other adaptive-biasing techniques like RF
D. Kwon, J. Lim, D. Oh, and T. B. Cho are with Samsung Semiconductor, envelope-tracking. In Section III, the core building blocks of the
Hwaseong-Si 445-330, South Korea (e-mail: [email protected]; proposed dynamically biased transmitter are described in detail
[email protected]; [email protected]; tb.cho@samsung. and design considerations are provided. Section IV presents the
com).
Color versions of one or more of the figures in this paper are available online
detailed measurement results of the realized prototype, and the
at http://ieeexplore.ieee.org. performance is compared with recent state-of-the-art transmit-
Digital Object Identifier 10.1109/JSSC.2015.2510023 ters. Finally, concluding remarks are provided in Section V.
0018-9200 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE JOURNAL OF SOLID-STATE CIRCUITS

processing blocks in baseband. As shown in the plots of the

PDF of the envelope amplitude at RF and the instantaneous
swing at BB in Fig. 3, and in the time-domain plot of a rep-
resentative 4G LTE20 signal at baseband and in RF in Fig. 4,
the individual I and Q baseband signal streams have more sig-
nal swing variation than the RF envelope. Effectively, each I
and Q stream has a larger PAPR (+3 dB) compared to the RF
envelope. At the same time, the rate of change of the amplitude
is slower in each I and Q stream. This implies that wherever
possible, dynamically biasing the baseband signal processing
blocks is more advantageous.
In our proposed dynamically biased power-mixer-based
transmitter architecture, the amplified and filtered baseband I
Fig. 1. Traditional power-mixer-based TX architecture.
and Q currents at the output of the V2I converter are fed to the
active double-balanced mixer. As the individual I and Q base-
band blocks (the V2I converter input and output branches) are
dynamically biased, the total dc power consumption from the
II. DYNAMIC B IASING V ERSUS F IXED B IASING
VDD node connected to the balun center-tap is reduced, leading
Fig. 1 shows the general architecture for a power-mixer- to a very power-efficient design.
based transmitter. The baseband signal is received from a DAC
in the MODEM baseband IC (BBIC) as a current and is con-
verted to a voltage and buffered before being fed to a biquad III. DYNAMIC B IASING IN T RANSMITTER —D ESIGN
low-pass filter. A resistor then converts the voltage to a current AND I MPLEMENTATION
that is linearly amplified and provided to the upconverter stage.
A. TX Signal Path
In a class-A power mixer, the V2I input and output branches
are biased with constant currents. Such a fixed-biasing scheme Fig. 5 shows the TX block diagram. The TX can be recon-
is illustrated in Fig. 2 (left). The baseband 2G/3G/4G signals figured to support different modes and covers the LB (699–
processed by such transmitters have nonconstant amplitudes 915 MHz) and MB/HB (1.4–2.7 GHz) bands using only two
and thus a certain peak-to-average-power-ratio (PAPR). As we baluns. I+, I−, Q+, Q− currents are provided from a current
see from the corresponding time domain plots in the fixed- DAC as baseband inputs to the RFIC TX. There can be two
biasing case, the bias current Idc is kept high enough to ensure kinds of interfaces between the BBIC DAC and the RFIC TX. In
that when the input current has high swing (e.g., in region the default case, the current from the DAC is first converted to
A), the instantaneous transistor current does not get clipped. a voltage, then buffered and filtered by a voltage-mode analog
However, as the ac current swing reduces (region B), the high baseband low-pass Rauch filter that implements a second-order
Idc bias current leads to power inefficiency. As we go to higher Butterworth transfer function, and attenuates the DAC sam-
order modulation schemes, the PAPR goes up. Consequently, pling image and noise. The filter output is then applied to the
peak swings happen more infrequently, further reducing power dynamically biased V2I. The resistor branch, buffer, and ana-
efficiency. log baseband filter can also be bypassed and the current from
On the other hand, in a dynamic-biasing scheme illustrated the DAC can be fed directly to the V2I. The voltage swing
in Fig. 2 (right), the V2I input and output branches are always at the input of the V2I is fed to a dynamic-biasing block that
biased with only the requisite amount of bias current. The goal comprises of a hybrid envelope detector/full-wave rectifier cir-
is to make sure that the minimum value of the actual current cuit and an envelope transconductor to generate a dynamically
in each of these transistors is always barely above 0 to avoid changing bias current that is used to bias the input branch of
any clipping and distortion. With the same dynamic dc current the V2I. The output branch of the V2I feeds the active double-
injected in the differential p and n halves, the output differential balanced I/Q quadrature upconverter (one for LB and one for
current is still a linear, undistorted, and amplified version of the MB/HB), transforming the input BB signal to RF.
input differential current. Cascode/current-steering RF VGA stage and R2R ladder
Traditionally, such dynamic biasing has often been imple- provide gain attenuation while shielding the LVT thin-oxide
mented at RF in envelope-tracking power amplifiers (ET-PA) devices of mixer and V2I gm devices from high-voltage sup-
[10]. Such ET-PAs require an envelope-detector circuit, usually ply of 1.8 V and large voltage swing from the on-chip baluns.
implemented as a squarer circuit, and a fast-settling, low-ripple, The baluns transform 50 Ω loads to different RF loads with the
and linear dc–dc supply modulator [11]. Being a single-ended aid of impedance control circuits, and implement differential to
circuit, nonlinearities introduced by the ET action introduce single-ended conversion. The supply voltage can be switched
AM–AM and AM–PM distortion [12]. In addition, for good between 1.8 and 1.2 V through two PMOS switches that work
EVM and linearity, the phase path and envelope path delays closely with the impedance control circuits to enable the best
need to be well matched, requiring additional circuit effort [13]. power saving when Pout is backed-off, without compromising
While dynamic biasing in an ET-PA is applied to the I–Q the linearity. A 10 bit dc offset cancellation DAC is connected
combined and upconverted RF signal, our proposed dynamic- to the V2I virtual ground nodes to calibrate the LO leakage.
biasing scheme is applied to the individual I and Q signal A 78 dB gain range is achieved in the TX by first slicing
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SETH et al.: DYNAMICALLY BIASED MULTIBAND 2G/3G/4G CELLULAR TRANSMITTER 3

Fig. 2. Fixed biasing versus dynamic biasing.

Fig. 7 shows the pole positions and the power-spectral-

density of the different thermal noise sources. vn2 /Δf repre-
sents the voltage noise spectral density at the output of the
analog LPF, M represents the current mirror ratio, and the
transconductance of the transistors Min , Mout and the current
source connected to the Min transistor are gm , M gm , and αgm ,
respectively. The first passive pole p1 is usually placed just
beyond the channel bandwidth to attenuate the noise and har-
monic distortion components of the signal from the LPF. The
large loop gain of −Gm Ro gm R1 ensures that the node Vx is
a virtual ground and that the output current Iout is a linear
undistorted version of the input current Iin . The nonlinear char-
acteristic of the matched transistors Min and Mout mean that the
gate voltage Vg contains higher harmonics of the input signal.
As a consequence, the second passive pole p2 is usually placed
beyond the third harmonic of the input signal.
The V2I transfer function is given by (1), where the pole
positions p1 and p2 are mentioned in Fig. 7
Fig. 3. PDF of envelope amplitude and signal swing at RF versus at baseband. 1 M gm 1 1
Iout = Vin · · · s ·
R1a + R1b gm + Gm Ro R1 1 + p1 1 + ps2
1
the V2I output gm device and mixer and cascode devices by
1 1 1
8× (18 dB), subslicing the output gm device in the last slice ≈ Vin · ·M · · . (1)
by 2× (6 dB), followed by 36 dB from the current-steering R1a + R1b 1 + ps1 1 + ps2
RFVGA and 24 dB from the R2R ladder.
The noise spectral density at the output of the V2I at a
B. Voltage-to-Current (V2I) Converter Design and Analysis frequency ω is given by (2)
 
The voltage output from the analog low-pass filter is con- 2
Iout M 2 gm2
γ (1 + α)
verted into a current and linearly amplified by the V2I converter. ≈ 4kT · γM gm + 2 · R2 +
Δf ω
1 + p2 gm
This is achieved by first creating a virtual ground using a high- 2

gain op-amp in feedback, and connecting a resistor that converts  2

R1a 1 1
the output voltage of the active-RC LPF to a linear current, + R1a + · 2 · 2
R1b 2 (R
gm 1a + R1b ) 1 + ωp2
which is then amplified using current mirrors and fed to the 1
2
power mixer. The single-ended version of the V2I that lets us vn2 1 M
+ · 2 ·  . (2)
perform this signal conditioning is shown in Fig. 6 along-with Δf (R1a + R1b ) 1+ ω2
· 1+ ω2
the differential to single-ended op-amp schematic. The op-amp p21 p22

consists of a high-gain differential amplifier stage followed by

a source-follower stage that is used to generate the gate voltage Equation (2) shows that the noise from the LPF and the resis-
and drive the low impedance of the R2 − C2 pole. tor is amplified by M 2 /R12 but filtered by the two passive poles.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE JOURNAL OF SOLID-STATE CIRCUITS

Fig. 4. Time domain waveform for 4G LTE20 100 RB signal at RF and at baseband.

Fig. 5. Dynamically biased multiband transmitter architecture (only I-path in baseband section shown).

Fig. 6. V2I op-amp schematic.

Similarly the noise from the transistor Min is amplified by M 2 determined by how much we want to tradeoff noise for linear-
and filtered only the R2 − C2 passive pole p2 . ity in the transmitter. For good noise performance, we need to
In a given mode, the mirror ratio M and the bias current lower the value of gm . But we cannot lower gm by a lot, because
is fixed by the Pout requirement, and the choice of gm is it would mean a smaller device and an increase in flicker noise.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SETH et al.: DYNAMICALLY BIASED MULTIBAND 2G/3G/4G CELLULAR TRANSMITTER 5

Fig. 7. V2I noise analysis.

Fig. 8. 4G LTE full RB and 1 RB signals in time and frequency domain.

Note that to reduce the flicker noise contribution from the V2I is a part of a three-transistor stack in the power-mixer branch,
mirror transistors, longer channel length of 1 µm is used. this can lead to nonlinearity due to signal compression in the
Lowering gm also worsens linearity as follows: First, a lower power-mixer.
gm implies that for the same swing on the input current Iin ,
the swing on the gate voltage Vg is higher. This can be bad
for linearity in the maximum Pout case where a large swing on C. Hybrid Envelope Detector/Full-Wave Rectifier and
the Vg node may cause the V2I op-amp’s transistors to leave Envelope Transconductor
the saturation region. Second, a lower gm also implies larger In this section, the design of two core circuit blocks
overdrive voltage required in the transistor Mout . Because Mout which enable dynamic-biasing is described: 1) a hybrid
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE JOURNAL OF SOLID-STATE CIRCUITS

Fig. 9. Hybrid differential to single-ended envelope-detector/full-wave rectifier.

envelope detector/full-wave rectifier—converts the differential

V2I input voltage into a dynamic voltage and 2) envelope
transconductor—converts the dynamic voltage into a current to
bias the V2I transconductance branch. The dynamic nature of
the transmitter biasing allows the transmitter to respond to the
input signals with varying bandwidth and PAPR. Therefore,
a brief introduction highlighting the unique nature of the 4G
transmit signals is provided before proceeding to the circuit
details.
A common attribute of 2G (GMSK/8-PSK) and 3G
(WCDMA/HSPA+) cellular standards is that the spectrum of
the modulated signal is always centered about dc. However,
for efficient spectrum utilization, the notion of resource-block
(RB) is introduced in 4G-LTE. The transmit spectrum is divided
into multiple RB units; each RB is 180 kHz wide. For exam-
ple, LTE20 comprises of 100-RB. The LTE transmitter can
be configured to operate in Full-RB (RB = 100) or Single-
RB (RB = 1) mode. In Full-RB mode, the transmit power is
spread over a bandwidth of 18 MHz and the modulated signals Fig. 10. Vdyn waveforms for exact tracking and envelope tracking. (a) Full-
are centered around dc. A representative LTE20 full-RB sig- wave rectification (exact tracking), small C, large Idischarge . (b) Envelope
nal is shown in Fig. 8(a). In contrast, in the Single-RB mode, detection (envelope tracking), large C, small Idischarge .
the power is concentrated within a single-RB which could be
placed at any one of the 100 available RBs. Since the signal- Vdyn , the transistor MSW shuts OFF and the capacitor C gets
power could be concentrated in any RB, the narrowband signals discharged by the Idischarge current, lowering Vdyn . The high-
could appear at an offset from dc as shown in Fig. 8(c). loop gain ensures that the static difference between Vdyn and
For Full-RB signals, the linearity-metric is the adjacent chan- the peak of Vin,p or Vin,n remains small. Thus, if the dc voltage
nel leakage ratio (ACLR) as shown in Fig. 8(b). For the on Vin,p or Vin,n is 0.9 V, and they swing ±0.5V, then the Vdyn
Single-RB case, the signal effectively looks like a tone at the voltage swings between 0.9 and 1.4 V.
offset frequency (in this case 9 MHz) with an envelope mov- The value of the capacitance C and the discharge current
ing at roughly the bandwidth of the signal (180 kHz). The Idischarge determines the Idischarge /C rate at which the voltage
key linearity performance metric is then the third-order counter Vdyn gets discharged. The value of C and Idischarge thus deter-
intermodulation product CIM3 as shown in Fig. 8(d). The mines how fast Vdyn tracks the greater of the voltages Vin,p or
contrasting nature of the modulation signals in Full-RB and Vin,n . Using a small capacitance (0.1 pF) and a large discharge
Single-RB transmission necessitates configuring the transmitter current (6.25 µA), we can track the absolute value of the input
in two different modes—a high-efficiency mode for wideband signal giving us the exact tracking mode or the full-wave rec-
signals and a high-linearity mode for the narrowband signals. tification mode that is used for wideband signals like Full-RB
The schematic for the hybrid envelope detector/full-wave 4G LTE. The simulated waveforms for this case are shown in
rectifier circuit is shown in Fig. 9(a). The three transistors Fig. 10(a). Next, using a large capacitance (10 pF) and a small
M1,p , M1,n , and M2 are all sized the same. As shown, the dif- discharge current (1.5625 µA), we can track the envelope, giv-
ferential input structure comprising of the transistors M1,p and ing us the envelope tracking mode that is used for narrowband
M1,n can be effectively redrawn as a transistor M1
that is driven

signals like 1 RB 4G LTE. The simulated waveforms for this
by the larger of the two p and n voltages, i.e.,
Vin,p/n
. As case are shown in Fig. 10(b).
shown in Fig. 9(b), when the larger of the voltages Vin,p or Vin,n After the Vdyn voltage has been generated, the envelope infor-
is larger than Vdyn , then the charge current through MSW charges mation (the full-wave rectified envelope or the peak detected
the cap C up, raising Vdyn . Similarly, as shown in Fig. 9(c), envelope) is available in the voltage domain. After this, a
when the larger of the voltages Vin,p or Vin,n is smaller than source-degenerated differential amplifier with diode-connected
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SETH et al.: DYNAMICALLY BIASED MULTIBAND 2G/3G/4G CELLULAR TRANSMITTER 7

Fig. 12. V2I currents.

Fig. 13. TX micrograph.

The final transistor currents and the output differential cur-

rent for the two modes are shown in Fig. 12. In both cases,
dynamic biasing ensures that the minimum instantaneous tran-
sistor current is close to 0. Also, while the instantaneous
transistor currents look distorted, the final V2I differential out-
Fig. 11. Envelope transconductor. put current is a linear amplified version of the input differential
current.
The choice of envelope tracking for the case of the 1 RB
loads is used as a transconductor to convert this envelope infor- signal (180 kHz bandwidth narrowband signal at an offset of
mation into the current domain as shown in Fig. 11. The current 9 MHz) can now be better understood from the waveforms in
IT is then mirrored (the mirrored side is implemented as a bina- Fig. 12(b). While changing the bias current at the same rate as
rily weighted current DAC that switches the output current over the instantaneous signal swing on Vin,p and Vin,n by using the
a 4 bit 16 level range) to generate the final dynamic bias current exact-tracking mode would lead to the best power savings, dis-
of Idyn into the p and n sides of the V2I input branch. By chang- turbing the “zero-crossings” of the input ac signal will reduce
ing the Vbase voltage (generated using a voltage DAC), we can the V2I linearity, preventing it from meeting the stringent CIM3
change the intercept of the Idyn versus Vdyn curve, i.e., change requirement. The slightly longer delay in the path from the
the fixed offset in the Idyn current as shown in Fig. 11. Next, by voltage Vin,p and Vin,n to Idyn (t) through the dynamic biasing
changing the mirror ratio, we can change the slope of the Idyn block, compared to the direct path from the voltage Vin,p and
versus Venv curve, changing effectively the amplitude as well as Vin,n to Iac (t) through the resistor R1 , means that when Iac (t)
the fixed offset of the Idyn current. rises, Idyn (t) may not rise as fast. This will lead to clipping and
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE JOURNAL OF SOLID-STATE CIRCUITS

thus worse linearity. On the other hand, using envelope tracking TABLE I
ensures that the “zero-crossings” of the input ac signal are not M EASURED P ERFORMANCE S UMMARY
disturbed at the same fast rate at which the signal is moving
(i.e., 9 MHz), but rather at the slower rate at which the envelope
is moving. While this leads to reduced power savings, it still
helps the V2I achieve a good CIM3 performance while using a
dynamic bias current.

D. Delay Mismatch Between Main Signal and Biasing

Signal Paths
Delay mismatch between the main signal and the biasing sig-
nal paths can affect the linearity performance in dynamically
biased systems. When dynamic biasing is implemented in the
RF domain, the difference in the frequencies of the main RF sig-
nal and the biasing signal (coming from an envelope-tracking
circuit) is much larger compared to the dynamic biasing scheme
proposed in the BB section in this work. As an example, if
an LTE20 signal that has been upconverted to 900 MHz goes
through a dynamically biased ET PA, then the envelope sig-
nal has a bandwidth of around 36 MHz while the RF signal
is at 900 MHz. The wide difference in frequencies makes it
difficult to achieve delay matching. On the other hand if the
LTE20 signal goes through a dynamically biased baseband sig-
nal chain, the biasing current has a bandwidth of 18 MHz while
the main signal has a bandwidth of 9 MHz, a much smaller
difference.
As shown in Fig. 5, before combining at the V2I input tran-
sistor, the main signal current is filtered by the R1a − C1 − R1b
passive network (pole p1 ). But in order to limit the in-band
droop from this pole (that can impact the EVM), the pole
p1 is kept at around 2× the bandwidth of the signal. This
in turn minimizes the delay of the signal current through the
R1a − C1 − R1b network. In addition, the Rfilt − Cfilt network
shown in the envelope transconductor of Fig. 11 can be used to
delay the envelope signal, thereby giving us a knob to match the
delays between the main signal and the envelope signal path.

E. Impact of Dynamic Biasing on Noise and Power-Mixer

Linearity
Dynamic biasing results in a lower average dc current, and
this leads to a lower TX Pdc and better battery life. But dynamic
biasing also leads to better noise-to-signal ratio. As discussed in
Section III-B, the noise from the baseband is dominated by the
thermal noise of the V2I input and output branch transistors,
and this noise is proportional to the transconductance of these
devices. This transconductance reduces as dynamic biasing
IV. M EASUREMENT R ESULTS AND D ISCUSSIONS
reduces the average transistor bias currents, effectively reducing
the noise of these transistors, thus improving the noise-to-signal The transmitter has been implemented in a 28 nm CMOS
ratio. Additionally, a lower average dc current reduces the VDS technology and occupies a total area of 1.45 mm2 including the
requirement in the three transistors (cascode transistor, mixer RF baluns and port switches. The chip micrograph is shown in
device, V2I output branch transistor) that are stacked between Fig. 13. A detailed overview of the measured TX performance
the balun and ground. This implies that for the same voltage in 2G, 3G, and 4G operating modes is presented in Table I. In
swing across the balun, a lower average dc current leaves more 2G mode, in the low-band, the transmitter achieves an ORFS of
margin for these transistors to remain in saturation, improving less than −70 dBc at 400 kHz offset while delivering >8 dBm
the linearity in the power mixer. Pout in GMSK mode and 5.8 dBm Pout in the 8 PSK EDGE
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SETH et al.: DYNAMICALLY BIASED MULTIBAND 2G/3G/4G CELLULAR TRANSMITTER 9

Fig. 14. TX output spectrum and constellation diagram for 4G LTE20 with 64 QAM modulation in Band 7 (2.5 GHz).

Fig. 15. TX output spectrum for 4G LTE20 1 RB signal in Band 1 showing Fig. 16. TX output spectrum for 4G LTE10 1 RB signal in Band 13.
CIM3 and CIM5 performance.

consumption. With a combination of these techniques, the TX

mode with a very low-noise leakage in the Rx band at 20 MHz power consumption rapidly scales down as the TX Pout is
offset. With an HSPA+ modulated signal, the transmitter has reduced, showing better performance compared to the Pdc ver-
an ACLR of <−40 dBc in both low-band (B5) and mid-band sus Pout profile measured in [6]. The total current consumption
(B1) while delivering >5 dBm Pout . In the 4G mode, the TX is becomes constant when the output power is reduced below
able to give out >4 dBm Pout across all bands with better than −20 dBm. This is because once the V2I output gm device is
−40 dBc E-UTRA ACLR. The 3fLO harmonic at the TX output subsliced in half in the last power-mixer slice (giving a total
is measured to be −25 dBc. power-back-off of 24 dB), the dc current in the power mixer
Fig. 14 shows the constellation and spectrum when the TX is fixed. After that, the output power is scaled down using
sends out a 3.7 dBm Pout 4G LTE20 (100 RB) 64 QAM mod- the programmable current-steering RFVGA and the R2R based
ulated signal with 8.2 dB PAPR at 2.5 GHz in Band 7. Exact RFVGA. With such a gain control line up, the TX output power
tracking is used for this mode and the TX power consumption is measured to be controllable over a range of greater than 80 dB
is 144 mW. Fig. 15 shows that by using envelope tracking mode, in the 3G/4G modes.
the same signal chain can be configured to provide <−66 dBc The advantages associated with the use of dynamic biasing
CIM3 and <−70 dBc CIM5 in the Band 1 4G LTE20 1 RB versus fixed biasing can be appreciated from the two output
case. Similarly, in Band 13, good CIM3 performance enables spectra shown in Fig. 18. As an illustration, with a 7.6 dB
the TX to meet the stringent 769–775 MHz spurious emission PAPRRF QPSK modulated LTE20 signal transmitted at 2.5 GHz
specification as shown in Fig. 16. in Band 7, we see that with exact tracking mode in dynamic
In Fig. 17, the measured TX dc power consumption is plotted biasing, the TX is able to send out 4.8 dBm Pout with an
as a function of the TX Pout for different modes and bands in ACLR of −41 dBc while consuming a dc current of 72 mA,
3G HSPA + and 4G LTE. In 3G mode, the base-band filter is but with dynamic biasing turned OFF, the dc current increases
always bypassed, while in the 4G mode, the base-band filter is to 156 mA. The large dc current increases the VDS of the three
bypassed when the Pout is reduced 6 dB from maximum Pout transistor stack, compressing the peak swing across the balun
(4 out of 8 output slices turned OFF). At this point, in primary turns. This in turn reduces the Pout to 2.8 dBm from
all the modes, the power-mixer supply voltage is switched 4.8 dBm and also worsens the nonlinear terms, raising the
to 1.2 V from 1.8 V, further reducing the baseband power ALCR to −36 dBc. Additionally, the measured noise-to-signal
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE JOURNAL OF SOLID-STATE CIRCUITS

actual measured total TX current (which includes the fixed cur-

rent in the analog baseband filter, voltage buffer, V2I op-amp,
and the bias blocks) reduces from 156 to 72 mA when dynamic
biasing is turned ON, representing a 53% savings.
Finally, a comparison with the state of the art is provided in
Table II. We first compare the performance of our active-mixer-
based TX with previously reported passive-mixer-based TX
architectures. Compared to the 25% duty cycle LO and passive
mixer architecture of [2] that supports only LB and MB, our
TX shows better linearity and power efficiency for wideband
signal performance (3G HSPA+, 4G LTE full-RB) while sup-
porting all three bands (LB/MB/HB). The narrowband (1 RB)
signal performance is not reported. In [3], a 33% duty cycle
LO driven passive-mixer-based TX achieves <−70 dBc CIM3
while delivering 2 dBm Pout in Band 13, and burns 144 mW in
the TX and 72 mW in the LO generation block. The relatively
high power consumption in the LO generation block results
from the need to generate and distribute 12 phases of 33% duty-
cycle clocks. These phases are generated by using a divide-by-2
and a divide-by-3 block. In the dynamically biased TX, only
four phases of 50% duty cycle generated using a divide-by-4
block are used, and consequently, for Band 13, the TX achieves
−63.5 dBc CIM3 while delivering 2.9 dBm Pout , and burns
177 mW in the TX and only 11 mW in the LO generation block.
In Band 1 at 1.95 GHz, the measured power consumption in our
Fig. 17. Measured TX dc power consumption versus RF output power for
TX is 154 mW in the TX and 20 mW in the LO generation block
different modes.
while delivering 1.4 dBm Pout with a CIM3 of −66.5 dBc.
While [3] does not report the performance and power consump-
tion numbers at Band 1 1.95 GHz, it is expected that using a
similar approach of generating 12 phases with 33% duty-cycle
will lead to a very high power consumption required in generat-
ing and distributing the 3fLO (∼6 GHz) signal from the PLL to
the TX and then generating and distributing the 12 phases of the
1.95 GHz signal among the multiple RF segments in the TX.
We next compare our work to other active-mixer-based TX
architectures. As shown in Table II, compared to the class-A
biased power mixers in [5] and [8], the dynamically biased
power mixer in our work provides better linearity while deliver-
ing higher Pout and burning lesser dc power. The performance of
the dynamically biased TX is compared to the class-AB biased
power mixer [6], [7] in Table II and Fig. 17. While the class-AB
TX utilizes four baluns to support LB and MB, in our dynam-
ically biased TX a single balun is used to support the MB and
HB bands (1.4–2.7 GHz). In the 3G mode, the dynamically
biased TX shows better power-efficiency than the class-AB TX
as shown in Fig. 17. In 4G, the class-AB TX consumes 101 mW
Fig. 18. TX output spectra with and without dynamic biasing.
while transmitting a 4.0 dBm LTE20 signal in B1, while the
dynamically biased TX consumes 143 mW while transmitting
ratio in the Rx band with 0 dBm Pout in these two cases worsens 4.1 dBm LTE20 signal in B1. The higher Pdc in the dynamically
from −159.8 dBc/Hz to −158 dBc/Hz when dynamic bias- biased TX can be attributed to the higher PAPR of the LTE20
ing is turned OFF, a result of increase in the gm of the main signal (7.6 dB versus 4.5 dB). As shown in Fig. 17, as the TX
noise-generating V2I current mirror transistors. Pout is backed off, the dynamically biased power mixer shows
As shown in the appendix, the theoretical percentage power better power-efficiency than the class-AB TX.
savings with an ideal exact-tracking
 based dynamically
 biased Note that for the same power-mixer current swing, the Class-
V2I section is given by 1 − 10−PAPRBB /20 ×100 where AB V2I converter requires twice the current mirror ratio as
PAPRBB = PAPRRF + 3. For the case shown in Fig. 18, with the dynamically biased V2I, worsening its noise performance.
PAPRBB = 10.6 dB, the theoretical expected percentage power In addition, the bias current in the Class-A/B V2I converter
savings in the dynamically biased section should be 70%. The is always changed at the same rate as the signal, and for a
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SETH et al.: DYNAMICALLY BIASED MULTIBAND 2G/3G/4G CELLULAR TRANSMITTER 11

TABLE II
M EASURED P ERFORMANCE C OMPARISON

1 Estimated from Fig. 9.2.6 (5).

1 RB signal centered at 9 MHz offset, this can be detrimen- A PPENDIX

tal in achieving <−57 dBc CIM3. On the other hand, in the T HEORETICAL P OWER S AVINGS W ITH DYNAMIC B IASING
dynamically biased V2I, when processing the 1 RB signal, the Assuming that the average bias current in the input branch
bias current is modulated at the rate of the signal’s envelope (the
in dynamic biasing and fixed biasing case shown in Fig. 2 are
rate of which is close to the signal bandwidth, around 180 kHz), Iaverage,dyn and Iaverage,fixed , respectively, and further assuming
allowing us to save power as well as achieve <−65 dBc CIM3.
that we use exact tracking where the bias current is the full-
wave rectified version of the input ac current Iac (t), and the
minimum instantaneous current through the transistor is 0, we
V. C ONCLUSION can write
In this paper, a complete SAW-less transmitter for 2G/3G/4G T
1
applications implemented in 28 nm standard CMOS process Iaverage,fixed = max {|Iac (t)|} and Iaverage,dyn = |Iac (t)| dt.
is presented. The power-mixer bias current is dynamically T
0
adjusted based on the instantaneous baseband signal swing (A1)
using a fully differential hybrid full-wave rectifier/envelope
detector circuit. Specific to the 4G case, the degree and nature Furthermore, the percentage savings can be written as
of the dynamic biasing current is adjusted based on the num- follows:
ber and offset of the RBs, to provide good ACLR as well  
as CIM3 performance. Leveraging these techniques the TX Iaverage,dyn
% Saving = 1 − × 100
achieves better power efficiency while transmitting large PAPR Iaverage,fixed
4G LTE signals compared to the previously reported active-  
1
T
and passive-mixer-based transmitters. A lower average bias-
T|Iac (t)| dt
0
ing current in the V2I output stage/power mixer branch fur- = 1− × 100. (A2)
max {|Iac (t)|}
ther reduces the out-of-band emission and improves linearity.
The proposed baseband dynamic biasing technique can also
The baseband PAPR (dB) of Iac (t) can be written as
be applied to other base-station/mobile transmitter blocks for
WLAN/Bluetooth/WiMax applications, where there is a need to  
2
achieve high performance with low power consumption while [max {|Iac (t)|}]
PAPRBB = 10 log10  (A3)
1 T 2
processing high PAPR signals. T 0 |Iac (t)| dt
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE JOURNAL OF SOLID-STATE CIRCUITS

which can be rearranged to write [7] P. Rossi et al., “An LTE transmitter using a class-A/B power mixer,”
in IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, Feb. 2013,
  pp. 340–342.
1 T 2
T 0 |Iac (t)| dt [8] T. Georgantas et al., “A 13 mm2 40 nm multiband GSM/EDGE/
= 10−PAPRBB /20 . (A4) HSPA+/TDSCDMA/LTE transceiver,” in IEEE Int. Solid-State Circuits
[max {|Iac (t)|}] Conf. Dig. Tech. Papers, 2015, pp. 160–161.
[9] S. Seth et al., “A dynamically-biased 2G/3G/4G multi-band transmitter
The two integrals in (A1) and (A4) can be written in their with >4 dBmPout , <−65 dBc CIM3 and <−157 dBc/Hz out-of-band
summation forms as follows: noise in 28 nm CMOS,” in IEEE Radio Freq. Integr. Circuits (RFIC)
Symp. Dig. Papers, 2015, pp. 147–150.
T N T [10] F. H. Raab et al., “Power amplifiers and transmitters for RF and
1 1  1 2 Microwave,” IEEE Trans. Microw. Theory Techn., vol. 50, no. 3, pp. 814–
|Iac (t)| dt = |Iac (tn )| and |Iac (t)| dt 826, Mar. 2002.
T N n=1 T
0 0 [11] J. Kim et al., “Highly efficient RF transmitter over broad average power
range using multilevel envelope-tracking power amplifier,” IEEE Trans.
N
1  2 Circuits Syst. I, Reg. Papers, vol. 62, no. 6, pp. 1648–1657, Jun. 2015.
= |Iac (tn )| . (A5) [12] K. Moon et al., “Investigation of intermodulation distortion of enve-
N n=1 lope tracking power amplifier for linearity improvement”, IEEE Trans.
Microw. Theory Techn., vol. 63, no. 4, pp. 1324–1333, Apr. 2015.
The Cauchy–Schwarz inequality [14] states that [13] J. Jeong et al., “Time mismatch effect in linearity of hybrid envelope
tracking power amplifier,” IEEE Microw. Wireless Compon. Lett., vol. 25,
 N
2  N
  N
 no. 8, pp. 550–552, Aug. 2015.
   [14] J. M. Steele, The Cauchy-Schwarz Master Class, 1st ed. Cambridge,
a n bn ≤ a2n · b2n . (A6) U.K.: Cambridge Univ. Press, 2004.
n=1 n=1 n=1 [15] 3GPP TS 36.101 Version 11.4.0 Release 11 Technical Specification,
[Online.] Available: http://www.etsi.org/deliver/etsi_ts/136100_136199/
Writing an = |Iac (tn )| and bn = 1 in (A6), we get 136101/11.04.00_60/ts_136101v110400p.pdf
N 2  N   N 
  2
 Siddharth Seth received the B.Tech. degree in elec-
|Iac (tn )| · 1 ≤ |Iac (tn )| · 1 tronics and electrical communication engineering
n=1 n=1 n=1 from the Indian Institute of Technology, Kharagpur,
 India, in 2006, and the M.S. and Ph.D. degrees

1 
N 1  N in electrical engineering from Stanford University,
|Iac (tn )| ≤ 
2
i.e., |Iac (tn )| . (A7) Stanford, CA, USA, in 2008 and 2013, respectively.
N n=1 N n=1 He has held internship positions at Microsoft
Research, Redmond, WA, USA; National
Semiconductor, Santa Clara, CA, USA; Broadcom
Then, from (A4), and by substituting (A5) in (A7), we get Corporation, San Jose, CA, USA; and LitePoint

1 T
Corporation, Sunnyvale, CA, USA. Since 2013,
T 0 |Iac (t)| dt
he has been with the Modem RF Laboratory, Samsung Semiconductor, San
≤ 10−PAPRBB /20 . (A8) Jose, CA, USA, where he is currently a Staff Engineer working on next
[max {|Iac (t)|}] generation cellular chipsets. His research interests are design and optimization
of analog/RF circuits and systems.
Finally, from (A2) and (A8) Dr. Seth was the recipient of the Dr. B. C. Roy Gold Medal for best all round
performance at IIT Kharagpur (2006), and the Robert Bosch Stanford Graduate
 Fellowship at Stanford University (2006).
% saving ≥ 1 − 10−PAPRBB /20 × 100. (A9)
Dae Hyun Kwon received the B.S. degree in elec-
tronics engineering from Korea University, Seoul,
ACKNOWLEDGMENT South Korea, in 2002, the M.S. degree in electri-
cal engineering and computer science from Seoul
The authors would like to thank Q. Zhou for help with the National University, Seoul, South Korea, in 2004,
proof of the inequality shown in (A7). and the Ph.D. degree in electrical and computer
engineering from the University of Illinois at Urbana-
Champaign, Champaign, IL, USA, in 2010.
From 2010 to 2013, he was with Broadcom Corp.,
R EFERENCES San Diego, CA, USA, where he designed RF trans-
mitters with on-chip power amplifiers for WLAN
[1] H. Xin and J. van Sinderen, “A low-power, low-EVM, SAW-less products. Since 2013, he has been with Samsung Electronics, Hwaseong, South
WCDMA transmitter using direct quadrature voltage modulation,” IEEE Korea, where he has worked on cellular RF transceivers.
J. Solid-State Circuits, vol. 44, no. 12, pp. 3448–3458, Dec. 2009.
[2] V. Giannini et al., “A multiband LTE SAW-less modulator with dBc/Hz
RX-band noise in 40 nm LP CMOS,” in IEEE Int. Solid-State Circuits Sriramkumar Venugopalan received the B.Sc.
Conf. Dig. Tech. Papers, 2011, pp. 374–376. degree from the Indian Institute of Technology (IIT),
[3] Y.-H. Chen et al., “An LTE SAW-less transmitter using 33% duty-cycle Kanpur, India, in 2008, the M.Sc. and Ph.D. degrees
LO signals for harmonic suppression,” in IEEE Int. Solid-State Circuits in electrical engineering from the University of
Conf. Dig. Tech. Papers, 2015, pp. 172–173. California, Berkeley, CA, USA, in 2009 and 2013
[4] B. Mohammadi et al., “A Rel-12 2G/3G/LTE-Advanced 2CC respectively.
Transmitter,” in IEEE Radio Freq. Integr. Circuits (RFIC) Symp. Dig. While at Berkeley, he worked in the BSIM Group
Papers, 2015, pp. 159–162. and pursued research and development of multigate
[5] O. Oliaei et al., “A multiband multimode transmitter without driver transistor compact SPICE models that contributed to
amplifier”, in IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, the industry standard BSIM-CMG model. Currently,
2012, pp. 164–165. he is with Samsung Electronics pursuing RF inte-
[6] N. Codega et al., “A current-mode, low out-of-band noise LTE transmit- grated circuit design in advanced semiconductor technology nodes. He has
ter with a class-A/B power mixer,” IEEE J. Solid-State Circuits, vol. 49, authored and coauthored more than 30 research papers in the area of semi-
no. 7, pp. 1627–1638, Jul. 2014. conductor device SPICE models and integrated circuit design.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

SETH et al.: DYNAMICALLY BIASED MULTIBAND 2G/3G/4G CELLULAR TRANSMITTER 13

Sang Won Son (S’96–M’02) received the B.S., M.S., Jaehyun Lim received the B.S. degree in electron-
and Ph.D. degrees in electrical engineering and com- ics engineering from Korea University, Seoul, South
puter science from the University of California at Korea, in 2001, the M.S. degree in electrical engineer-
Berkeley, Berkeley, CA, USA, in 1995, 1998, and ing, and the Ph.D. degree in computer science and
2002, respectively. engineering from the Pennsylvania State University,
From 2002 to 2010, he was with Marvell State College, PA, USA, in 2004 and 2009, respec-
Technology Group, Santa Clara, CA, USA, where tively.
he worked on multiple generations of WLAN Since 2009, he has been with Samsung Electronics,
transceiver products. From 2010 to 2012, he was a Sr. Hwaseong, South Korea, where he is working on
Department Manager with Mediatek in San Jose, CA, cellular RF transceivers.
USA, where he developed RF and analog circuits for
cellular and connectivity applications. He is currently a Senior Director with
Samsung Semiconductor, Inc. San Jose, CA, USA, where he is developing the
next generation cellular chipset solutions. Dongjin Oh received the B.S. and M.S. degrees
in electrical engineering and computer science from
Ajou University, Suwon, South Korea, in 2009 and
2011, respectively.
Yongrong Zuo received the B.S. and M.S. degrees From 2011 to 2015, he was a Research Engineer
in electronic engineering from Tsinghua University, with System LSI Business, Samsung Electronics,
Beijing, China, in 1994 and 1997, respectively, and Hwaseong, South Korea, where his research focused
the Ph.D. degree in electrical engineering from the on CMOS RF/analog integrated circuits and RF
University of Delaware, Newark, DE, USA, in 2006. transceivers for wireless communication.
From 2006 to 2013, he was with Qualcomm,
Boxborough, MA, USA, as a Senior Engineer, and
later a Staff Engineer on RF and analog IC design
Thomas Byunghak Cho (S’90–M’95–SM’14)
for various wireless applications. Since 2013, he has
received the B.S. degree from the University of
been with Samsung Semiconductor Inc., San Jose,
California, Los Angeles, CA, USA, in 1989, and
CA, USA, as a Senior Staff Engineer, designing low-
the M.S. and Ph.D. degrees from the University of
power and low-voltage RF and analog ICs using advanced deep-submicron
California, Berkeley, CA, USA, in 1991 and 1995,
CMOS technologies.
respectively, all in electrical engineering.
From 1995 to 1996, he was a Postdoctoral
Researcher with the University of California,
Venumadhav Bhagavatula (S’11-M’15) received Berkeley, CA, USA. From 1996 to 2000, he was
the B.E. degree in electronics and communication with Level One Communications, San Francisco, CA,
from the University of Delhi, New Delhi, India, the USA, developing CMOS RF transceiver products
M.Tech. degree in electronic design technology from for cordless phone applications. In 2000, he cofounded Wireless Interface
the Indian Institute of Science, Bangalore, India, Technologies, Dublin, CA, USA, developing CMOS RF transceivers for
and the Ph.D. degree in electrical engineering from WPAN/WLAN applications, and the company was later acquired by Chrontel,
the University of Washington, Seattle, WA, USA, in San Jose, CA, USA. In 2004, he joined Marvell Semiconductor, Santa Clara,
2005, 2007, and 2013, respectively. CA, USA, developing CMOS RF and analog IC products for various wired and
Since 2014, he has been with the Modem Lab, wireless connectivity applications. Since 2012, he has been the VP of the RF
Samsung Semiconductors Inc., San Jose, CA, USA. Development Team, Samsung Electronics, Hwaseong, South Korea, focusing
His research interests include RF/mm-wave and low- on the development of CMOS RF/Analog ICs and data converter IPs for mul-
power mixed-signal circuits. timode multiband cellular modem applications. His research interests include
Dr. Bhagavatula was the recipient of the CEDT Design Medal from the CMOS analog integrated circuits for high-speed analog-to-digital interfaces and
Indian Institute of Science (2007), the Analog Devices Outstanding Student wireless communication systems.
Designer Award (2012), and the corecipient of the Best Student Paper Award at Dr. Cho was the corecipient of ISSCC Jack Kilby Award for Outstanding
the RFIC 2014. Student Paper in 1997.