Institutionen för systemteknik
Department of Electrical Engineering
Examensarbete
Low-Power Low-Noise IQ Modulator Designs in
90nm CMOS for GSM/EDGE/WCDMA/LTE
Examensarbete utfört i Radioelektronik vid Tekniska högskolan i Linköping
av
Mattias Johansson, Jonas Ehrs LiTH-ISY-EX--10/4330--SE
Linköping 2010
Department of Electrical Engineering Linköpings tekniska högskola Linköpings universitet Linköpings universitet SE-581 83 Linköping, Sweden 581 83 Linköping
Low-Power Low-Noise IQ Modulator Designs in
90nm CMOS for GSM/EDGE/WCDMA/LTE
Examensarbete utfört i Radioelektronik
vid Tekniska högskolan i Linköping
av
Mattias Johansson, Jonas Ehrs LiTH-ISY-EX--10/4330--SE
Handledare: Magnus Nilsson
ST-Ericsson
Henrik Fredriksson
ST-Ericsson
Niklas Karlsson
ST-Ericsson
Examinator: Atila Alvandpour
ISY, Linköpings universitet
Avdelning, Institution Division, Department
Division of Electrical Engineering Department of Electrical Engineering Linköpings universitet
SE-581 83 Linköping, Sweden
Datum Date 2010-03-19 Språk Language Svenska/Swedish Engelska/English ⊠ Rapporttyp Report category Licentiatavhandling Examensarbete C-uppsats D-uppsats Övrig rapport ⊠
URL för elektronisk version
http://www.ek.isy.liu.se http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-54552 ISBN — ISRN LiTH-ISY-EX--10/4330--SE Serietitel och serienummer Title of series, numbering
ISSN —
Titel
Title Effekt- och Brus-Effektiva IQ Modulatorer i 90nm CMOS för GSM/EDGE/WCD-MA/LTE Low-Power Low-Noise IQ Modulator Designs in 90nm CMOS for GSM/EDGE/WCDMA/LTE
Författare
Author Mattias Johansson, Jonas Ehrs Sammanfattning
Abstract
The current consumption of the IQ modulator is a significant part of the total current consumption of a mobile transmitter platform and reducing it is of great interest. Also, as the WCDMA/LTE standards specifies full duplex transmissions and Tx and Rx are most often using the same antenna, it is crucial to have a solution with low noise generation. Two new proposals have been studied with the aim to reduce the current consumption and noise contribution of the IQ modulator. A current mode envelope tracking IQM is the first of the studied designs. This implementation lowers the bias currents in the circuit in relation to the amplitude of the baseband input signals, meaning that a low input amplitude results in a lowering of the current consumption. It proves to be very efficient for baseband signals with a high peak-to-average ratio. Simulations and calculations have shown that an average current reduction of 56 % can be achieved for an arbitrary LTE baseband signal.
The second is an entirely new passive mixer design where the baseband voltages are sequentially copied to the RF node, removing the need for V-to-I conversion in the mixer which reduces current consumption and noise. Results from simulations has proven that this design is fully capable of improving both current consumption as well as the noise levels. With an output power of 4.0 dBm, the power consump-tion was 43.3 mW, including clock generating circuits. This, combined with the fact that the design is small and simple, means that there is definitely a possibility to replace the present IQM design with a passive mixer.
Nyckelord
Keywords RF, ASIC, Analog Integrated Circuit, IQ Modulator, IQM, CMOS, 90nm, Predis-tortion, Mixer, Envelope Tracking, Direct Conversion
Abstract
The current consumption of the IQ modulator is a significant part of the total current consumption of a mobile transmitter platform and reducing it is of great interest. Also, as the WCDMA/LTE standards specifies full duplex transmissions and Tx and Rx are most often using the same antenna, it is crucial to have a solution with low noise generation. Two new proposals have been studied with the aim to reduce the current consumption and noise contribution of the IQ modulator. A current mode envelope tracking IQM is the first of the studied designs. This implementation lowers the bias currents in the circuit in relation to the amplitude of the baseband input signals, meaning that a low input amplitude results in a lowering of the current consumption. It proves to be very efficient for baseband signals with a high peak-to-average ratio. Simulations and calculations have shown that an average current reduction of 56 % can be achieved for an arbitrary LTE baseband signal.
The second is an entirely new passive mixer design where the baseband voltages are sequentially copied to the RF node, removing the need for V-to-I conversion in the mixer which reduces current consumption and noise. Results from simulations has proven that this design is fully capable of improving both current consumption as well as the noise levels. With an output power of 4.0 dBm, the power consump-tion was 43.3 mW, including clock generating circuits. This, combined with the fact that the design is small and simple, means that there is definitely a possibility to replace the present IQM design with a passive mixer.
Acknowledgments
We want to express our gratitude to the ST-Ericsson supervisors of the thesis project, Magnus Nilsson, Henrik Fredriksson and Niklas Karlsson and to other staff members of the RF ASIC department at ST-Ericsson who have been helpful with all the different problems we have encountered. Thanks goes to the exam-iner of the thesis, Atila Alvandpour and to our opponents, Mattias Johansson and Richard Kjerstadius, for improving the quality of the thesis and the report. Fi-nally, we thank our girlfriends, Charlotta and Anna, who had to endure 6 months of distance relationship and see us leave for Lund every week.
Thanks for making this thesis work possible!
List of Abbreviations
AM Amplitude Modulation BER Bit Error Rate
BPSK Binary Phase Shift Keying
CMOS Complementary Metal Oxide Semiconductor DAC Digital to Analog Converter
EDGE Enhanced Data rates for Global Evolution FM Frequency Modulation
FWR Full-Wave Rectifier
GMSK Gaussian Minimum Shift Keying GPRS General Packet Radio Service
GSM Global System for Mobile communications
HB High Band
HSDPA High Speed Downlink Packet Access HSPA High Speed Packet Access
HSUPA High Speed Uplink Packet Access IM InterModulation
IMD InterModulation Distortion IQ In-phase Quadrature phase
IQM In-phase Quadrature phase Modulator ISSCC International Solid-State Circuits Conference
LB Low Band
LO Local Oscillator ix
LTE Long Term Evolution
MOSFET Metal Oxide Semiconductor Field Effect Transistor NMOS N-type Metal Oxide Semiconductor
OFDM Orthogonal Frequency-Division Multiplexing OP OPerational amplifier
OQPSK Offset Quadrature Phase Shift Keying PA Power Amplifier
PAR Peak to Average Ratio PM Phase Modulation
PMOS P-type Metal Oxide Semiconductor PSK Phase Shift Keying
QAM Quadrature Amplitude Modulation QPSK Quadrature Phase Shift Keying RF Radio Frequency
Rx Receiver
SAW Surface Acoustic Wave
SC-FDMA Suppressed Carrier Frequency Division Multiple Access SNR Signal to Noise Ratio
SSB Single Side Band
Tx Transmitter
VGA Variable Gain Amplifier
Contents
1 Introduction 1 1.1 Purpose . . . 1 1.2 Goals . . . 2 1.3 Limitations . . . 2 1.4 Simulation environment . . . 2 1.5 Document outline . . . 3 2 Theory 5 2.1 Analog modulation . . . 5 2.1.1 Amplitude modulation . . . 6 2.1.2 Frequency modulation . . . 6 2.1.3 Phase modulation . . . 7 2.2 IQ data . . . 7 2.3 IQ modulator . . . 8 2.4 Digital modulation . . . 9 2.4.1 PSK . . . 10 2.4.2 GMSK . . . 10 2.4.3 QAM . . . 11 2.5 PAR . . . 122.6 Wireless communication standards . . . 12
2.6.1 GSM (2G) . . . 13 2.6.2 GPRS (2.5G) . . . 13 2.6.3 EDGE (2.75G) . . . 13 2.6.4 WCDMA (3G) . . . 13 2.6.5 HSPA (3.5G) . . . 14 2.6.6 LTE (4G) . . . 14 2.6.7 Technical comparisons . . . 14 2.7 Transmitter efficiency . . . 17 2.8 Noise . . . 17 2.8.1 Thermal noise . . . 17 2.8.2 Flicker noise . . . 17
2.8.3 MOSFET noise model . . . 18
2.9 Non-linearity . . . 18
2.9.1 IM3 . . . 20 xi
2.10 Predistortion . . . 20 2.11 SAW-filter . . . 21 3 Reference IQM 23 3.1 Environment . . . 23 3.2 Architecture . . . 24 3.2.1 Overview . . . 24 3.2.2 Amplifier . . . 26 3.2.3 Mixer . . . 28 3.2.4 RC filter . . . 31
3.3 Simulations and results . . . 33
3.3.1 Testbench setup . . . 33
3.3.2 Power consumption . . . 34
3.3.3 Linearity . . . 35
3.3.4 Noise . . . 36
4 Current Mode Envelope Tracking IQM 37 4.1 Theory . . . 37
4.2 Implementation . . . 38
4.2.1 Rectifier . . . 39
4.2.2 RC filter . . . 42
4.3 Simulations and results . . . 45
4.3.1 Testbench setup . . . 45
4.3.2 Power consumption . . . 47
4.3.3 Linearity . . . 48
4.3.4 Noise . . . 53
5 Direct Voltage IQM 55 5.1 Theory . . . 55
5.2 Implementation . . . 58
5.2.1 Output stage noise analysis . . . 58
5.2.2 Mixer dimensioning . . . 59
5.3 Simulations and results . . . 60
5.3.1 Testbench setup . . . 60
5.3.2 Power consumption . . . 62
5.3.3 Linearity . . . 62
5.3.4 Noise . . . 63
5.3.5 Non-overlapping clock signals . . . 64
5.3.6 Simulating with 4PGFD clocking circuit . . . 66
5.3.7 Discussion . . . 67
6 Predistorted Direct Voltage IQM 69 6.1 Theory . . . 69
6.2 Implementation . . . 70
6.3 Simulations and results . . . 71
6.3.1 Testbench setup . . . 71
Contents xiii
6.3.3 Linearity . . . 72
6.3.4 Noise . . . 73
6.3.5 Simulating with 4PGFD clocking circuit . . . 75
7 Conclusions 77 7.1 Future work . . . 78
7.1.1 Envelope Tracking IQM . . . 78
7.1.2 Direct Voltage IQM . . . 80
7.2 Final words . . . 81
Chapter 1
Introduction
The continued enhancement of wireless technology puts great pressure on the companies acting in the field. In the mobile phone industry one can see a global scurry towards being the first to present the latest technology on the market. More and more applications and functions are squeezed into the phones to meet the requirements of the customers. While increasing the size of the phones extensively is not an option, more area and power efficient batteries and circuits is the key for obtaining the capabilities needed for new and better functionalities.
In this thesis we examine the IQ modulator for a mobile communication chip, trying to find new ways to reduce the current consumption. Also, as the present implementation at ST-Ericsson of the IQ modulator, the Original IQM, has a somewhat high noise floor, it is followed by an expensive SAW (Surface Acoustic Wave) filter off-chip to reduce the Tx-Rx interference. Constructing an IQ modu-lator that lowers this noise floor would mean that it might be possible to remove this expensive filter and reduce cost and area. The new IQ modulators are de-signed to be able to handle signals from several wireless communication standards, where the focus for this thesis lies on GSM, EDGE, WCDMA and LTE.
1.1
Purpose
The thesis was carried out at ST-Ericsson in Lund, Sweden, with the main focus to analyze and reduce the current consumption of the IQ modulator (see Section 2.3), which is a common part in the Tx chain of an RF circuit. The analysis involved examining two new IQ modulators and understand their behavioral and performance.
1.2
Goals
The goals for this thesis are:
• Study and understand the limitations of the following IQ modulator designs: – Current Mode Envelope Tracking IQ Modulator
– Direct Voltage IQ Modulator
• Compare these two modulators to each other as well as to a slightly modified version of the Original IQM, called the Reference IQM, in the context of power consumption, linearity and noise.
1.3
Limitations
• This thesis is concluded on a schematics hierarchy level, not taking circuit layout complications into concern.
• The IQ modulators in this thesis are designed to work with several mobile communication standards and several frequency bands, spanning an output frequency range of 700-2500 MHz. However, this thesis is limited to studies at 2000 MHz.
• To adjust the gain of the Original IQM between 1 dB and 20 dB using 1 dB steps, several select signals can be set to logarithmically enable or disable 70 identical mixer units whose currents adds together to achieve the wanted gain. Throughout the thesis, this variable gain functionality is not implemented. Instead, simulations have been done using always maximum gain. This equals to enabling all 70 mixer units concurrently.
1.4
Simulation environment
All circuits were constructed in 90 nm CMOS process using Cadence Virtuoso Suite and were simulated using Spectre. Coherently throughout the simulations a single tone SSB modulated sinusoidal was used as baseband signal. SSB means that the Q-part of the signal has the same frequency but phase shiftedπ
2 from the I-part as (see Section 2.2 for IQ theory):
I(t) = cos(ω0t) (1.1)
Q(t) = cos(ω0t −
π
1.5 Document outline 3 These signals linearily mixed in an IQ modulator (Section 2.3), as in Equation 2.9, results in the RF signal:
sRF(t) =I(t) cos(ωct) + Q(t) sin(ωct)
= cos(ω0t) cos(ωct) + sin(ω0t) sin(ωct) (1.3)
=cos[(ωc− ω0)t] 2 + cos[(ωc+ ω0)t] 2 + cos[(ωc− ω0)t] 2 − cos[(ωc+ ω0)t] 2 = = cos[(ωc− ω0)t] (1.4)
The reason for using an SSB modulated sine wave is that many of the specifications provided by ST-Ericsson are referred to when using such input. Also, an SSB sine wave is an easy test case because in a linear system it resolves in only one output frequency. In simulations, a carrier frequency of 2 GHz was used to cover the most commonly used frequency bands of modern communication standards. The applied baseband signal at 10 MHz refers to the case of an LTE signal with a 20 MHz bandwidth, creating a wanted frequency at 1990 MHz.
1.5
Document outline
The thesis report is separated into seven chapters and is written for Master of Science students already familiar with basic signal and electronics theory. Chapter 2 covers some basic RF theory sections, including modulation techniques and the functionality of a theoretical IQ modulator. The following four chapters, all covers one implemented design each, where Chapter 3 describes a reduced version of the Original IQMdesign, referred to as the Reference IQM. Following in chapters 4-6 are the new implemented ideas, which are called the Envelope Tracking IQM, Direct Voltage IQM and Predistorted Direct Voltage IQM respectively. The last chapter presents the conclusions and proposed future work.
Chapter 2
Theory
This chapter tries to cover the necessary background theory that is needed for the intended reader to understand the thesis. It includes basic RF theory as analog and digital modulation as well as descriptions of IQ data and the IQ modulator, as well as some brief descriptions of some of the mobile communication standards within the focus of this thesis. There is also some notes about PAR, transmitter efficiency and circuit noise and in the end there are short discussions about linearity and predistortion.
2.1
Analog modulation
The basic principle behind signal modulation is transmitting information by the modification of a typically high frequency carrier wave
sc(t) = Accos(ωct + ϕc) (2.1)
where ωc= 2πfc and fcis the carrier frequency. This carrier and its notation will
be used as an example throughout the chapter. This chapter will also use a simple input signal, called the modulating signal, sm(t):
sm(t) = Amcos(ωmt + ϕm) (2.2)
where, similar as above, ωm= 2πfmand fmis the modulating signal’s frequency.
Sending information using modulation has several advantages. It allows many transmitters to transmit at the same time by letting every transmission have their own frequency (or frequency band). It is also easier to construct antennas that are efficient for higher frequencies than lower, and the higher frequency antennas can be built smaller as well. [3]
All signal modulation schemes are based on three modulation modes; Ampli-tude, Frequency and Phase Modulation. [3]
2.1.1
Amplitude modulation
AM is the simplest modulation scheme and was also the first to be used. It is based on inserting information in the carrier wave’s amplitude. The modulated AM signal could mathematically be described as: [3]
sAM(t) = Ac[1 + sm(t)] cos ωct (2.3) Time A m p li tu d e
Figure 2.1. Example of AM. sm(t)superimposed in red.
2.1.2
Frequency modulation
As far as AM is simple to understand and is rather simple to implement it still has it’s drawbacks. The biggest disadvantages is the susceptability to noise and poor power efficiency. The concept of frequency modulation tries to avoid this by letting the FM signal not depend on its amplitude, but on its frequency. The modulated FM signal is given by: [3]
sF M(t) = Accos(ωct + 2πf∆ t
Z
0
sm(τ )dτ ) (2.4)
where f∆ is the frequency deviation, which is the maximum frequency change of
the carrier wave in one direction. [3]
Time A m p li tu d e
2.2 IQ data 7
2.1.3
Phase modulation
The third modulation mode is phase modulation, where the phase of the car-rier wave is changed proportionally to the modulating wave’s amplitude. Phase modulation is given by: [3]
sP M(t) = Accos(ωct + sm(t)) (2.5)
Important to notice is that FM and PM are using the same physical property, the angle of sc(t), and therefore they cannot be used simultaneously. This is important
to recognize as it will be discussed later.
Time A m p li tu d e
Figure 2.3. Example of PM. sm(t)superimposed in red.
2.2
IQ data
It is common in radio communications to discuss the transmitted signal using the two-dimensional IQ representation, where I stands for In-phase and Q for Quadra-ture phase. To explain this representation we first look closer at the modulating wave:
sm(t) = Amcos(ωmt + ϕm
| {z } θm
) (2.6)
At any given moment in time, this signal could be represented as a phasor in a two-dimensional plane (the IQ-plane) using polar coordinates where Am is the
amplitude and θmis the angle as shown in Figure 2.4. Looking at that figure it is
pretty straightforward to explain IQ which is merely the cartesian coordinates of this representation, i.e.
Id= Amcos(θm) (2.7)
Qd= Amsin(θm) (2.8)
However, now one would maybe wonder why going the extra step and transform the perfectly simple polar representation to IQ representation. The answer lies in the performance of IQ modulator hardware and simplicity in digital signal processing. Constructing a precise phase shifter for high frequencies is difficult, it is easier to
Figure 2.4. Modulating wave as a phasor in the IQ-plane
only adjust the amplitudes of I and Q instead. Also, as we will see in Section 2.4, several digital modulation schemes operate directly on the I and Q components of the carrier signal which without an IQ modulator leads to unnecessary conversion in the digital domain from IQ to polar representation. [10]
2.3
IQ modulator
In the IQM the I part of the baseband signal is multiplied with a cosine carrier wave and the Q part is multiplied with a 90◦ shifted cosine carrier wave, or in
other words, a sine carrier wave. Carriers separated by 90◦ are called quadrature
carriers, or orthogonal carriers. As these two signals are orthogonal they can be added together without information loss, resulting in the outgoing RF signal. The transmitted signal is expressed as: [10]
sRF(t) = I(t) cos(ωct) + Q(t) sin(ωct) (2.9)
2.4 Digital modulation 9 To demodulate this signal, sRF is again multiplied with the same carrier waves
and I and Q can be retreived after lowpass filtering. For example, to retreive the I part again the following multiplication is done
sRF(t) cos(ωct) = I(t) cos2(ωct) + Q(t) cos(ωct) sin(ωct) =
= I(t) 2 +
1
2[I(t) cos(2ωct) + Q(t) sin(2ωct)] (2.10) and then lowpass filtering gives I(t)
2 .
Figure 2.6. IQ demodulator
As understood, the IQM is not a modulator associated with a certain modulation scheme, it can actually be used with any modulation scheme. The actual mod-ulation is typically made in the digital domain before being transmitted to the IQM.
2.4
Digital modulation
In digital modulation the information to be sent are digital bits that are com-bined into symbols representing one or more bits. These symbols are commonly transmitted using a constant symbol rate defined by the communication standard in use. For example, the symbol rate in GSM is 270.833 kilosymbols/sec. Each symbol consists of 1 bit which gives a total data rate of 270.833 kbps for speech, data and control signals. This is divided between 8 mobile stations as GSM uses time-division access. [21]
Frequency bandwidth today is restricted by law in most countries. The gov-ernments decide what frequencies are to be allocated for different uses. Therefore a high bandwidth efficiency is wanted when designing wireless communication standards. Different modulation techniques with higher and higher bandwidth ef-ficiency have been developed over the years to cope with the growing usage of the frequency spectrum. Some of these are described in the following sections.
(a) 2-PSK (b) QPSK (c) 8-PSK
Figure 2.7. Popular PSK constellations
2.4.1
PSK
PSK (Phase Shift Keying) is a digital modulation scheme that modulates the phase of the carrier to transmit digital data. The number of possible phase positions defined in PSK can be arbitrary. For example, with 2-PSK there are two possible phases defined and a transmitter can therefore transmit one bit per symbol, by modulating the carrier as: [17]
Acsin(2πfct) , representing digital ’1’ (2.11)
Acsin(2πfct + π) , representing digital ’0’ (2.12)
Of course, a 2-logarithmic number of possible phases, or states, gives the possibility to fully encode binary data in these states.
By looking at the IQ plane for PSK we can see the good usage of IQ-data. PSK is using a phase shifting algorithm which is easily realized by changing the I part and the Q part of the signal. In figure 2.7 2-PSK, 4-PSK (also called QPSK) and 8-PSK constellations are plotted in the IQ-plane. By using a higher order PSK more bits can be transmitted per symbol but it is of course more sensible to interference as noise causes the received signal to not exactly point at a valid constellation position. [3, 17]
Worth mentioning is the OQPSK, which is a QPSK modulation scheme with a time offset between the I and the Q part of half a symbol. By using regular QPSK there is always the possibility of a 180◦ phase transition when going from 00 to
11, for example. With an offset the I and Q part changes value at different times which leads to a maximum phase transition of 90◦not causing the envelope to go
to 0 for a short time. This increases the bandwidth efficiency. [13]
2.4.2
GMSK
Another way of representing binary data is to let the direction of a phase shift represent either ’0’ or ’1’ and hold the amplitude of the IQ phasor at a constant
2.4 Digital modulation 11 level. GMSK (Gaussian Minimum Shift Keying) is a modulation technique that uses this concept when encoding data. The GMSK signal is similar to the one produced by OQPSK. The difference is that each one of the square shaped pulses in OQPSK is replaced with a gaussian filter shaped pulse. Low side lobes and a narrower main lobe is obtained when using the gaussian filter instead of a rect-angular pulse, which results in less interference between frequency channels. The frequency response and impulse response of the gaussian filter are described by equation 2.13 and 2.14 respectively. [15, 16, 25]
H(f ) = e−(f /a)2 (2.13)
h(t) =√πae−(πat)2 (2.14)
The constant a determines the 3dB bandwidth B of the filter according to equation 2.15. [25]
B = a q
ln√2 (2.15)
2.4.3
QAM
As said in Section 2.2, there are two variables that can be altered in the carrier wave, the amplitude and the angle. This is not made use of in PSK modulation as it is only the angle that is modulated. QAM (Quadrature Amplitude Modulation) however, takes advantage of both degrees as it spans up a quadratic constellation diagram where both the amplitude and the angle are variables. Analogous to M-PSK, M here denotes the possible states in the constellation; in common radio traffic ranging from 4-QAM (identical to QPSK) to 64-QAM. In 64-QAM every symbol consists of 6 bits, therefore increasing data throughput. [3, 13, 17]
Interesting to mention is that the M-QAM modulation is especially efficient with the IQM as it spans up a quadratic constellation. This means that the states in the constellation are equidistant and has a maximum possible distance between them, which leads to as little interference as possible between states. As the interference is low, the signal power can be lowered as well. To clarify one can think of the state positions for a 64-PSK constellation, where the states would be very close to each other resulting in high interference, therefore the M-QAM constellation is more efficient.
(a) 16-QAM (b) 64-QAM
Figure 2.8. Popular QAM constellations
2.5
PAR
The definition of the PAR (Peak to Average Ratio) of a signal x is P AR = P (x)ˆ¯
P (x) (2.16)
where ˆP (x) is the peak power and ¯P (x) is the average power of the signal x. A high PAR means that there are large power changes in the signal and it sets high demands on especially amplifiers in the circuit. If the PAR is low then the amplifier does not need high linearity constraints but if the signal has a high PAR then either the signal must be attenuated which reduces the output power or the power amplifier must have a large dynamic range which leads to high power consumption. [7]
For example, the PAR of a sine wave x(t) = A sin(wt) is P ARsin= ˆ P (x) ¯ P (x)= A2 A2/2 = 2 = 10 log(2) dB = 3.01 dB (2.17)
Some of the modulation schemes have a low probability for producing high ampli-tude spikes. When calculating the PAR value one can limit the signal samples to for example 99% of the total samples. By removing the uppermost percentage of the samples, possible spikes will be neglected. This is referred to as PAR0.99.
2.6
Wireless communication standards
This section shortly describes the wireless communication standards from the sec-ond generation to the fourth. If the reader is interested in more information regarding these standards, the authors refers to the references of each subsection. The last subsection compares GSM, EDGE, WCDMA and LTE regarding some interesting parameters. GPRS is excluded because it uses the same modulation
2.6 Wireless communication standards 13 as GSM. HSPA is basically an earlier form of LTE and is therefore also excluded. The main difference for the purpose of this thesis is that LTE can use 64-QAM in the uplink, which is from mobile to base station, and HSPA can not.
2.6.1
GSM (2G)
A certain band in the frequency spectrum is allocated for the GSM standard to be used for communication between mobile stations and base stations. Within this spectrum there are 124 discrete frequency carriers defined, where each car-rier is partitioned into 8 time-divided slots. In total this gives 124 ∗ 8 = 992 available traffic channels within each radio cell to be distributed among users. When transmitting data over these channels, GSM uses the modulation technique GMSK. [23]
2.6.2
GPRS (2.5G)
A big step in the evolution towards the third generation of telecommunication standards was GPRS, often referred to as 2.5G to show the link between 2G and 3G. GPRS introduces a packet-switched network, which still uses the GSM network nodes along with some additional GPRS specific nodes, to handle the transport of packet data. The modulation technique used for GPRS is GMSK, the same as for GSM. [4, 20]
2.6.3
EDGE (2.75G)
Further improvement of the GSM technology was made when implementing EDGE. By introducing the modulation technique 8-PSK, EDGE is able to transfer 3 bits per symbol compared to the 1 bit per symbol of the GMSK technique used by GSM and GPRS. This leads to a higher data throughtput even though the band-width is the same. A drawback is that 8-PSK is more sensitive to noise due to the shorter distance between the states in the constellation diagram (see Figure 2.7). To solve this problem, EDGE switches from 8-PSK to GMSK when the SNR becomes to low, e.g. when the user is located near the edge of a radiocell. [18]
2.6.4
WCDMA (3G)
To achieve a high data throughput WCDMA (Wideband Code Division Multi-ple Access) uses a wide frequency bandwidth where the user data is spread by multiplying it with quasi-random bits which are derived from orthogonal CDMA spreading codes. By adding these CDMA codes to the signals, it is possible for multiple users to transmit at the same frequency at the same time.
Just as GSM, WCDMA uses a combination of time and frequency division with a 5 MHz separation between the carrier frequencies. For modulation of the uplink data WCDMA uses QPSK. [8]
2.6.5
HSPA (3.5G)
HSDPA is a protocol for transfering data from the base station to the mobile station. Among other improvements over WCDMA it uses adaptive modulation that initially uses QPSK but if the signal quality is good it can improve data rates by using 16-QAM or 64-QAM. [5]
In addition to the downlink protocol, HSUPA was released to manage the uplink traffic and as a pair they are commonly called HSPA. HSUPA is using the same adaptive modulation technique as HSDPA, except that it can not use 64-QAM. Some extra overhead information about e.g. power control and scheduling are embedded in the data sent from the mobile station to the base station, as one wants to minimize the amount of calculations made in the mobile. This information is handled by the base station instead. [5]
2.6.6
LTE (4G)
In the downlink LTE uses another access technique called OFDM to increase the spectral efficiency. With this, the information in an assigned band is divided onto several subcarriers, 15 kHz apart, all orthogonal to each other. This increases spectral efficiency as the subcarriers can be placed close to each other. In the uplink, to reduce the PAR, a similar technique called SC-FDMA is used, which also uses orthogonal subcarriers but transmits the data sequentially instead of in parallel. [9]
As in WCDMA/HSPA different modulation techniques are used in the uplink depending on the condition of the link, but the choice of modulation techniques is wider. LTE can switch between QPSK, 16-QAM and 64-QAM. The bandwidth of LTE can vary between 1.4 and 20 MHz. [1, 19]
2.6.7
Technical comparisons
The interesting information about the different mobile standards for this thesis is the value of the PAR, channel bandwidth and what modulation techniques are used. The PAR value gives a hint of which modulation techniques will benefit the most from the Envelope Tracking IQM and the channel bandwidth sets the requirement in the common mode loop of this IQM design. More information is available in Section 4.3.4. Due to the possibility for WCDMA and LTE to generate high amplitude spikes, 1% of the highest values have been removed from the signal, which will give the PAR0.99 value for these signals.
MatLab scripts provided by ST-Ericcson were used to generate baseband sig-nals of the different standards and the PAR values were calculated from the ob-tained signals. In Table 2.1, PAR I/Q is the PAR value for the separate I or Q baseband signals, which in this case are equal, and PAR I+Q is for the combined signal. The scripts generate signals with some random deviation from the ideal values to imitate the reality.
To get a good view over the amplitude levels of the different signals, graphs showing the probability of the amplitude were created. These can be seen in Figure 2.9, where the blue line represents the probability distribution over the
2.6 Wireless communication standards 15 Standard Modulation PAR I/Q PAR I+Q Channel bandwidth
GSM GMSK 3.03 0.00 200 kHz
EDGE 8-PSK 6.24 3.29 200 kHz
WCDMA QPSK 5.03 3.54 5 MHz
LTE 64-QAM 8.39 5.39 20 MHz
Table 2.1. Technical data of wireless standards [1, 2, 6]
amplitude and the dashed red line shows the cumulative probability. The graphs show the probability distributions for the I baseband signals, which have the same amplitude probabilities as the Q baseband signals, with a minimum amplitude of 0 V and a maximum of 300 mV.
P ro b a b il it y (% ) Amplitude mV GSM (GMSK) C u m u la ti v e P ro b . (% ) 0 50 100 150 200 250 3000 50 100 0 1.1 2.2 P ro b a b il it y (% ) Amplitude mV EDGE (8-PSK) C u m u la ti v e P ro b . (% ) 0 50 100 150 200 250 3000 50 100 0 1.1 2.2 P ro b a b il it y (% ) Amplitude mV WCDMA (QPSK) C u m u la ti v e P ro b . (% ) 0 50 100 150 200 250 3000 50 100 0 1.1 2.2 P ro b a b il it y (% ) Amplitude mV LTE (64-QAM) C u m u la ti v e P ro b . (% ) 0 50 100 150 200 250 3000 50 100 0 1.1 2.2
2.7 Transmitter efficiency 17
2.7
Transmitter efficiency
A value that is useful when doing circuit comparisons in RF with the aim to decrease power consumption is the transmitter efficiency. It is defined as
Transmitter efficiency = Transmitted power
Consumed DC power (2.18) The value tells us how much of the input DC power is actually fed to the antenna. The power not fed to the antenna generates unnecessary heat, which is important to reduce as it must be transported away with expensive heatsink solutions. [3]
2.8
Noise
Noise in electrical circuits can be divided into two groups, interference noise and inherent noise. Interference noise is noise originating from other sources than the observed circuit, such as noise on a power supply line or electromagnetic interfer-ence from other circuits nearby. Inherent noise is the noise that the circuit itself generates. This noise is a variable that could be reduced by proper schematic and layout design. The following sections describes the main causes of inherent noise in a MOSFET transistor. [11]
A definition of a noise source’s voltage spectral density would also be in or-der. The voltage spectral density function, V2
n(f ), or In2(f ), is a function of the
average power at a certain frequency. Fact is that many noise sources are fre-quency dependent so this is crucial for calculations of the total noise at a certain bandwidth. [11]
2.8.1
Thermal noise
Thermal noise is generated by the thermal excitation of the charged carriers in a conductor. The noise is proportional to the absolute temperature of the conductor and is independent of the frequency, i.e. the spectral density is constant. [11]
For example, the noise in a resistor is primarily thermal noise and its voltage spectral function is:
Vr2(f ) = 4kBT R (2.19)
where kB is Boltzmann’s constant, T is absolute temperature and R is the
resis-tance. [11]
2.8.2
Flicker noise
Flicker noise is also called 1/f noise because it is inverse proportional to the frequency. It usually comes from ’traps’ in the component where carriers get stuck for some period of time and then is released. Its spectral density can be written:
Vn2(f ) =
k2 v
f (2.20)
Figure 2.10. Noise model of a MOSFET in saturation
2.8.3
MOSFET noise model
The inherent noise from a MOSFET is mainly thermal noise and flicker noise. The thermal noise is easy to understand as it directly originates from the resistive channel of a MOSFET. In ohmic mode the thermal noise is easily described by
Vd2(f ) = 4kBT rDS ⇒ Id2(f ) =
4kBT
rDS
(2.21) where rDS is channel resistance. However, when the transistor enters active mode
the channel cannot be described homogenous any longer. The thermal noise is then more accurately approximated by: [11]
Id2(f ) = 4kBT
2 3
gm (2.22)
The flicker noise in a MOSFET is modeled as a voltage source connected in series with the gate:
Vg2(f ) =
K
W LCoxf (2.23)
where K is dependent on physical characteristics of the device, W is transistor width, L length and Cox is gate capacitance per unit area. Worth noting is that
p-channel transistors have typically less flicker noise because the carriers (holes) are less likely to be trapped. In Figure 2.10 the total noise model of a MOSFET in active mode is shown. [11]
2.9
Non-linearity
A non-linear transfer function in an RF transmitter degrades not only its own spectra but also the adjacent channels, interfering others transmissions. Unfor-tunately, all electrical systems are somewhat non-linear because of the physical limitations of the components. Thus, for accurate calculations the system needs to be modeled using a non-linear model. The output of a non-linear system can be modeled as an infinte polynomial expression,
y(t) = α1x(t) + α2x2(t) + α3x3(t) + ... (2.24)
2.9 Non-linearity 19 The property of non-linearity in a circuit can be observed in different ways. One way is the occurrence of harmonics in the output signal when having a periodic input signal. One easy example is to let the input signal to a third order system be x(t) = cos(ωt), then the output signal will be
y(t) = α1cos(ωt) + α2cos2(ωt) + α3cos3(ωt)
=α2 2 + (α1+ 3α3 4 ) cos(ωt) + α2cos(2ωt) 2 + α3cos(3ωt) 4 (2.25)
As seen these harmonics can be found at multiples of the input frequency and therefore they can rather easily be attenuated using filtering.
Another way to discover and measure non-linearities is to measure the in-termodulation products. Inin-termodulation products occur when the input signal contains more than one frequency. These multiple frequencies now modulates with themselves, simplest seen with an input of x(t) = cos(ω1t) + cos(ω2t), again in a
third order system:
y(t) =α1[cos(ω1t) + cos(ω2t)] +
α2[cos(ω1t) + cos(ω2t)]2+
α3[cos(ω1t) + cos(ω2t)]3 (2.26)
This expression expands to the expression visualized by the following table: Angular frequency Amplitude
0 α2 ω1 α1 + 94α3 ω2 α1 + 94α3 2ω1 12α2 2ω2 12α2 ω1± ω2 α2 2ω1± ω2 34α3 2ω2± ω1 34α3 3ω1 14α3 3ω2 14α3
Table 2.2. Intermodulation products in a third order system
When ω1 and ω2 are close to each other in frequency the problem with
inter-modulation are the components 2ω1− ω2 and 2ω2− ω1, which are close to the
input frequencies and thus hard to filter out using filtering. The calculations also reveals that the main problem is the third order non-linearity, the second order only creates unwanted components at much higher (or lower) frequencies that are easier to filter out. In fact, when calculating with higher order systems than
three it can be seen that all even order non-linearites creates components that are easy to filter out. That means that the intermodulation problem is the odd order non-linearities. [14]
2.9.1
IM3
In this thesis the measure for non-linearities is the third order intermodulation distortion, IM3 (or IMD3) of an SSB baseband signal. With a single tone simula-tion in an IQ modulator as this, the IM3 is attained from a harmonic folding back to the in-band such as:
• First the third harmonic of the LO signal, 3ωc, is mixed with the fundamental
tone in the baseband signal, ωm, creating the unwanted harmonic signal
3ωc+ ωm(the 3ωc− ωmis not created because the third harmonic of the LO
signal is phase shifted 180◦).
• In the mixing process also the wanted frequency is created, ωc− ωm.
• These two components are now intermodulated, effectively folding back a component, (3ωc+ ωm) − 2(ωc− ωm) = ωc+ 3ωm. This occurs in-band and
is very important to reduce with an as linear system as possible.
2.10
Predistortion
Predistortion refers to the deliberate introduction of a non-linearity - a predistor-tion - to a signal, before it is being applied to a system, with the aim to compensate for the systems own non-linearities. The transfer function of the predistortion must then be as close as possible to the opposite of the transfer function of the system. To illustrate, an amplifier with a transfer characteristic of the one in Figure 2.11(a) can be compensated with a predistortion device with the opposite characteristics, as in Figure 2.11(b). These two non-linearities cancels each other out and the out-put signal is linear to the inout-put signal despite going through a non-linear amplifier. This technique is common in high power amplifier design. [12]
Input O u tp u t 0 1 2 3 4 5 6 7 0 10 20 30 40 50 60 70 80
(a) Non-linear amplifier
Input O u tp u t 0 1 2 3 4 5 6 7 -70 -60 -50 -40 -30 -20 -10 0 10 (b) Predistortion circuit Input O u tp u t 0 1 2 3 4 5 6 7 0 2 4 6 8 10 12 14
(c) Sum of both systems
2.11 SAW-filter 21
Figure 2.12. NMOS current mirror
The concept of predistortion is used in all IQ modulators in this thesis, except the one in Chapter 5. All predistortions in these circuits can be seen as ordinary current mirrors as in Figure 2.12. As known, a current mirror working in saturation mode is a non-linear transfer from current to voltage approximately originating from the equation
iin= 1 2Kp W1 L1 (VGS1− Vt1)2(1 + λVDS1) (2.27)
where Kp is a technology dependent constant, W1 and L1 is width and length of
the N1 transistor and λ is the channel-length modulation parameter. The same way there is a non-linear transfer from voltage to current from the same formula:
iout = 1 2Kp W2 L2 (VGS2− Vt2)2(1 + λVDS2) (2.28)
As VGS are equal in both transistors and as both transistors have the same device
characteristics (same Kp and Vt), the transfer function from iinto iout is:
iout iin = W2L1 W1L2 VGS2− Vt2 VGS1− Vt1 2 1 + λV DS2 1 + λVDS1 = = W2L1 W1L2 1 + λVDS2 1 + λVDS1 (2.29) Which, if VDS1= VDS2 is assumed, gives the now linear transfer function
iout iin =W2L1 W1L2 (2.30)
2.11
SAW-filter
The SAW filter is a common component in RF circuits due to its high selectivity. It is often used as a band pass filter for high frequencies. SAW stands for Surface Acoustic Wave and the name is explained by the function of the filter. The electri-cal signal is converted into a mechanielectri-cal wave using a co electri-called IDT (Interdigital Transducer) and is then propagated as an acoustic wave through a piezoelectric substrate before being converted back to an electrical signal. As the substrate is contructed to have a preferred direction of magnetic orientation, it, together with
the IDT, has a frequency response that can be tuned by constructing the IDT differently. These filters are however relativly large and costly, so removing these is always wanted. [22]
Chapter 3
Reference IQM
The Original IQM on-chip is divided into two practically identical blocks, one for the LB (Low Band) frequencies (824 to 915 MHz) and one for the HB (High Band) (1710 to 1980 MHz). As mentioned in Section 1.3, this thesis does not study the full bandwidth and therefore only the HB part of the IQM is presented here. However, if a design is working satisfactory at higher frequencies it is assumed that it will also work for lower frequencies. More specifically, the carrier frequency used is 2 GHz with a baseband frequency of 10 MHz creating a wanted output signal at 1990 MHz, as mentioned in Section 1.4.
The Reference IQM circuit is a stripped and slightly modified version of the Original IQM due to problems with cadence PSS convergence and extensive sim-ulation times. The changes are briefly described in the list below and more infor-mation can be found in the architecture section.
• The clock buffer and frequency divider are replaced with ideal sources. • The OP in the amplifier circuit is replaced with an ideal Verilog-A version
and the gain is set to 100.
• Capacitors have been added to the RC filter between the amplifier and mixer. The simplifications result in a circuit with better performance concerning linearity, current consumption and noise which are the parameters examined in the thesis. However, as the Envelope Tracking IQM, presented in Chapter 4, uses the same simplifications it will not affect the comparison of the two circuits.
3.1
Environment
The Original IQM is embedded in the Tx part of the chip. It receives baseband I and Q signals from the digital domain, via DA converters. After mixing the base-band signals with the LO frequency, the RF signal is put through a VGA (Variable Gain Amplifier) stage. As transmitting and receiving is done simultaneously in WCDMA/LTE and the closest Rx channel is only 45 MHz higher than Tx, there
Figure 3.1. Original IQM environment
is a need for keeping the Tx to Rx interference as low as possible. As there is only one antenna, the Tx signal is connected to the Rx signal via a duplexer, which is a combination of two SAW filters. One of the filters is centered around the Tx frequency and one around Rx. The filter around the Tx frequency has the task to reduce the noise within the Rx band of the Tx signal and the other filter eliminates the Tx signal from the Rx input. Due to problems for the duplexer to fully remove the Tx noise, an additional SAW filter is located between the VGA and the PA stages, ahead of the duplexer. The SAW filter are used for its good selectivity, but it is also expensive. If the IQM had better SNR it could be removed.
3.2
Architecture
3.2.1
Overview
An overview of the Original IQM can be seen in Figure 3.2. There are four sub-circuits in the IQM: a clock buffer, a frequency divider, an amplifier and a mixer. The I signal and the Q signal have their own amplifier and mixer, as seen in the figure. The input resistance Rin is set to control the gain of the amplifier, which is a negative feedback OP amplifier described below. In Figure 3.3, the resistors with resistance R between the amplifier and the mixer construct a low pass filter together with the Cc capacitors and the parasitic capacitance of the input stage
of the mixer. The function of this filter is to reduce the far-out noise from the amplifier stage which is the dominating noise in the circuit. With a better filter-ing here it may be possible to remove the Tx SAW filter. This RC filter is also described below, as well as the mixer, which output currents are added to create the RF signal.
The frequency divider is responsible of creating the quadrature LO clocking. However, for the purpose of the thesis, the mixer and amplifier are the most inter-esting parts. Therefore, and because of extensive simulation times, the clock buffer and frequency divider are excluded from this report and replaced with ideal volt-age sources in the testbenches. However, as a reference, the current consumption of these parts is simulated separately and is included in the power consumption calculations.
3.2 Architecture 25
Figure 3.2. Overview of Original IQM
Figure 3.3. Overview of Reference IQM
Name Value Rin 2 kΩ R 500 Ω Cc 7.73 pF
3.2.2
Amplifier
Figure 3.4. Reference IQM amplifier architecture
Transistor Width Length P1-P3 20 µm 400 nm N1-N3 3 µm 200 nm N4-N9 1.2 µm 100 nm
Table 3.2. Transistor dimensions of the amplifier
Name Type Unit Description
in+ Input A Positive baseband signal in− Input A Negative baseband signal CM Input V Common mode voltage for OP vpos1p8 Supply V High supply voltage i.e. 1.8 V vpos1p2 Supply V Low supply voltage i.e. 1.2 V out+ Output V Positive output
out− Output V Negative output
3.2 Architecture 27 The amplifier circuit in Figure 3.4 is a fully differential operational amplifier with a negative feedback over transistors N2 and N3. The feedback loop controls the output gain and keeps it at a wanted level. The common mode voltage for the OP is also used as a bias voltage applied at the gate of N1 creating a bias current, which is mirrored through the PMOS transistors at the top. As the transistors in the loop are designed to match the mixer (see Figure 3.5), this mirrored bias current is the maximum output current in the mixer stage. The amplifier stage is also a predistortion of the signal, increasing linearity, as described in Section 2.10. The OP used in the circuit is as stated an ideal component with a gain of 100. The Verilog-A code of the OP is included below. With this design, a common mode value of 960 mV gives a symmetrical input current swing through the input resistors (Rin), which is the task for the OP common mode voltage. The common mode of the OP has no relationship to the input common mode voltage except to get a symmetrical input current swing so that there is no DC current drawn from the input sources.
// V e r i l o g −A f o r OP
‘ i n c l u d e " c o n s t a n t s . vams" ‘ i n c l u d e " d i s c i p l i n e s . vams"
module TxLpfDiffOpamp ( Outn , Outp , Cm, Gnd , Vcc , Inn , Inp ) ; output Outn , Outp ;
i n p u t Inn , Inp ; i n o u t Vcc , Gnd ,Cm;
e l e c t r i c a l Outn , Outp , Inn , Inp , Gnd , Vcc ,Cm; r e a l sp [ 0 : 1 ] ;
parameter r e a l Gain = 1 0 0 ; parameter r e a l Rin = 12K; parameter r e a l Rout = 7 5 ; branch ( Inp , Inn ) In ; branch ( Outp , Outn ) Out ; a n a l o g b e g i n
I ( In ) <+ V( In ) / Rin ; V( Out ) <+ I ( Out ) ∗ Rout ;
sp [ 0 ] = −1.000000∗2∗ ‘M_PI∗1G; sp [ 1 ] = 0 ; V( Outp , Gnd) <+ l a p l a c e _ z p (V( In ) ∗ Gain , , sp )∗0.5+V(Cm) ; V( Outn , Gnd) <+ −1∗ l a p l a c e _ z p (V( In ) ∗ Gain , , sp )∗0.5+V(Cm) ; end endmodule
3.2.3
Mixer
The task for the mixer is to combine the amplified baseband signals with the LO frequency to obtain the wanted RF signal. This is done with the quadrature LO signals generated by the frequency divider. The frequency divider generates three differential signals: LO, divLOI and divLOQ. Inside the I part mixer (see Figure 3.5) the positive LO signal and the divLOI signal creates two 25% duty cycle clocks for the transistor columns, alternating between keeping column I2 and I3 active with keeping column I1 and I4 active. In the Q mixer, with the negative part of the LO signal and the divLOQ signal, the opposite two 25% duty cycle clocks are created. This means that, in the full system, there are always one column pair active at a time, alternating between the I part mixer and the Q part mixer. A description of how the duty cycles are created can be found in Figure 3.6. The ideal clock generating sources replacing the clock buffer and frequency divider are perfectly ideal with a voltage swing of 0 to 1.2 V and no rise or fall time.
Figure 3.5. Reference IQM mixer architecture (I-part)
Transistor Width Length N1-N4 3.12 µm 250 nm N5-N8 2.32 µm 200 nm N9-N16 960 nm 100 nm
3.2 Architecture 29 Name Type Unit Description
BBI+ Input V Positive amplified baseband signal BBI− Input V Negative amplified baseband signal
divLOI+ Clock V Positive In-phase frequency divided LO signal divLOI− Clock V Negative In-phase frequency divided LO signal LO+ Clock V In-phase LO signal
on Input V 1.8V ⇒ Enable, 0V ⇒ Disable vpos1p8 Supply V High supply voltage i.e. 1.8 V vpos1p2 Supply V Low supply voltage i.e. 1.2 V iout+ Output A Positive output
iout− Output A Negative output
Table 3.5. Inputs and outputs of the I-part mixer
(a) LO signals (b) Duty cycles
Figure 3.6.
The 25% duty cycle clocking in this circuit is not a conventional mixer design, it is more common to use 50% duty cycle. In Figure 3.7, the I part of a conventional mixer with 50% duty cycle is shown. The Q part is connected from the right in the figure. Using this clocking, all 4 columns (I and Q) are active at all time, consuming 2 times the current as in the mixer using 25% duty cycle. However, the conventional mixer has twice the output power as well. This can be seen in the conversion gain of the LO signals. The Fourier series of a square wave with a duty cycle of η is:
s(t) = η + 2 ∞ X n=1 sin(nπη) nπ cos(nω0t) (3.1)
By looking at the differential clock signal, where the differential clock is sdif f(t) =
s(t) − s(t + π), its Fourier series is: s(t) = 4 π ∞ X n=0 sin([2n + 1]πη) 2n + 1 cos([2n + 1]ω0t) (3.2) At the LO frequency (n=0) a conversion gain from each clock signal is 4
π at 50%
duty cycle (η = 0.5) and 4 √
2π at 25% duty cycle (η = 0.25), giving half the power
of a 50% duty cycle conversion.
In the Reference IQM design though, only two of the eight columns are active at a time, also halving the propagated noise. This means that the 25 % duty cycle mixer is consuming half the current, producing half the output power compared to the 50 % clocking but keeping the SNR constant.
3.2 Architecture 31
3.2.4
RC filter
Between the amplifier and mixer, an RC-filter is located with the task to reduce the noise of the differential output from the amplifier. The resistors along with the parasitic gate capacitances of the N5-N8 transistors in the mixer (see Figure 3.5) are the components of this filter. By adding extra capacitors in parallel with the parasitics, one can reduce the cutoff frequency of the filter and thereby suppress more noise. The total effective capacitance is represented by Ctwhich is
the physical capacitor combined with the parasitic capacitance. The theoretical architecture of the filter can be seen in figure 3.8, where A is the differential output from the amplifier, M is the input to the mixer, R is the value of the resistors and Ctsymbolizes the total effective capacitance as stated above.
(a) (b)
Figure 3.8. Theoretical RC filter circuits
Theoretically speaking there are 2 different filters, though with the same fre-quency response. The first one is a common mode filter that consists of 1 resistor and 1 parasitic capacitance added with the extra capacitor, represented by the red line in figure 3.8(a). There are 2 copies of this filter, one for each of the single ended signals. The filter has an RC constant according to equation 3.3 which comes from the serial coupled resistor and capacitance.
RCcm= RCt (3.3)
The second filter is a differential filter which consists of the 2 resistors and both total capacitances. Due to the ground connection of the capacitors, they have one mutual node, which implies a serial connection of the two Ct capacitances in a
differential view. The blue line in figure 3.8(b) shows the path which creates the differential filter. The calculation of the RC constant for the differential filter is shown below.
RCdif f = 2R
Ct
As can be seen in equations 3.3 and 3.4 the RC constants is the same for the differential and single ended filters (RCdif f = RCcm), which results in the same
cutoff frequency for both filters, described in equation 3.5. f3dB =
1 2πRCt
(3.5) The effective value of the parasitic capacitance was obtained by performing an AC analysis of the filter. With values for f3dB and R, one can easily calculate
Cp using equation 3.5. With R = 500 Ω, the parasitic capacitance was calculated
to Cp= 226 fF. To suppress noise but still make sure that possible 10 MHz LTE
baseband signals are forwarded, the wanted cutoff frequency was set to f3dB= 40
MHz for this thesis work. The filter can be unpredictable and unstable if only relying on the parasitics, therefore the added capacitors (Cc) should be larger than
Cp. The above values gives a Cc of 7.73 pF which results in the total capacitance
Ct= Cc+ Cp = 7.96 pF. The simulated frequency response of the filter is shown
in Figure 3.9. As stated in Section 1.3, all mixer components has a multiplicity of 70, hence the calculated Cp symbolizes the effective gate capacitance of 70 mixer
stages.
3.3 Simulations and results 33
3.3
Simulations and results
3.3.1
Testbench setup
The testbench of the IQM is shown in Figure 3.10. As previously discussed, the clock buffer and frequency divider is removed and replaced with ideal input signals. All input signals and their specifications can be found in Table 3.7. The load for the IQM circuit is a resonance load, working as a harmonic filter, with its resonance frequency tuned at the LO frequency at 2 GHz. The inductor’s and capacitor’s values are calculated from the ohmic load at RL= 175 Ω and the Q-value of 3.48
giving enough attenuation of the harmonics at 3 times the LO frequency. The Q-value, or the Quality factor, describes how good selectivity the filter has, and a higher Q-value means higher selectivity and better filter characteristics. Explicitly:
Q =RL r CT 2L = 3.48 f0= 1 2π√2LCT = 2 GHz RL=175 Ω (3.6)
Where CT equals the total capacitive load, which means CL combined with all
parasitic capacitances connected to the output nodes (CT = CL + Cp). The
equations in 3.6 gives the values L = 2 nH and CT = 1.58 pF. According to
simulations, a value of CL = 1.3 pF results in a harmonic filter with the peak at
f0= 2 GHz.
Figure 3.10. Testbench of Reference IQM
Name Value RL 175 Ω
CL 1.3 pF
L 2 nH
Name Type Signal level [min max] Unit I+ Input [600 1200] mV I− Input [600 1200] mV Q+ Input [600 1200] mV Q− Input [600 1200] mV CM Input 960 mV divLOI+ Clock [0.0 1.2] V divLOI− Clock [0.0 1.2] V LOI Clock [0.0 1.2] V divLOQ+ Clock [0.0 1.2] V divLOQ− Clock [0.0 1.2] V LOQ Clock [0.0 1.2] V vpos1p8 Supply 1.8 V vpos1p2 Supply 1.2 V
Table 3.7. Reference IQM testbench signals
3.3.2
Power consumption
The testbench has two voltage supplies, one at 1.2 V and one at 1.8 V, where in this testbench the 1.2 V is used for biasing some transistors in the amplifier circuit. The absolute majority (>99.999%) of the DC current is drawn from the 1.8 V supply, which generates the current and power consumptions presented in Table 3.8. The average output power with an SSB baseband signal at 10 MHz was in this case 5.5 dBm, giving a transmitter efficiency of 10.5 %.
Average current consumption 19.0 mA Average power consumption 34.2 mW Average output power 5.5 dBm Transmitter efficiency 10.5 %
Table 3.8. Results from the Reference IQM
To give a better idea of the power consumption of a complete IQM circuit, the current consumption of the LO generating circuits, i.e. the clock buffer and frequency divider, was simulated apart from the testbench and added to the above values. The resulting current and power consumption can be seen in Table 3.9, where the added overhead current consumption is 10.9 mA with a supply voltage of 1.2 V.
3.3 Simulations and results 35 Average current consumption @ 1.2 V 10.9 mA
Average current consumption @ 1.8 V 19.0 mA Average power consumption 53.8 mW Average output power 5.5 dBm Transmitter efficiency 7.6 %
Table 3.9. Results from the Reference IQM with added clock generation overhead
3.3.3
Linearity
The IM3 specification for the IQM is -37 dBc. However, due to the circuit mod-ifications with the ideal clocking and OP there is an improvement of linearity.
Relative IM3 distortion -43.5 dBc
Table 3.10. Result from the Reference IQM
3.3.4
Noise
The noise for the Reference IQM is presented in Figure 3.12. The relative noise spectrum density at 45 MHz offset (the nearest Rx channel) is found in the table below.
The presented results does not show the real truth about the noise levels, due to the many ideal devices as the OP and clock generation. However, they are presented here to make a comparison possible between this circuit and the Envelope Tracking IQM.
Relative noise spectrum density
at 45 MHz offset -169.9 dBc/Hz
Table 3.11. Noise result for the Reference IQM
Chapter 4
Current Mode Envelope
Tracking IQM
4.1
Theory
The basic idea of this version of the IQ modulator is to make the current consump-tion of the circuit to be somewhat proporconsump-tional to the input amplitude. Meaning that while the input amplitude is high, the IQM has a high current consumption, whilst a low input results in a lower consumption. This should result in a high benefit for signals with a high PAR value, such as LTE.
In the Reference IQM the P2 and P3 transistors of the amplifier (same as in Figure 4.3) serves as constant current sources. By substituting the common mode voltage of the OP with the envelope tracking signal (ET ), and replace the PMOS transistors with adjustable current sources which depend on the envelope of the baseband input signal, the current consumption of the amplifier will have a relation to the amplitude of the input. The envelope signal of a sinusoidal input can be seen in Figure 4.1 and Figure 4.2 shows the theoretical layout of the modified amplifier circuit where the NMOS transistors are the same as in the amplifier of the Reference IQM.
As the current through the adjustable current sources decreases, the current through the N2 and N3 transistors also has to be decreased to prevent the input nodes from dropping in voltage. The relationship of the currents in the input nodes is described by equation 4.1.
iN = iCS+ iIN (4.1)
The amplifier will draw a current (iIN) from the input signal when the current
(iCS) through the current sources decreases, to maintain the constant common
mode current through N2 and N3 (iN). This can cause problems for the buffers
in the DACs preceding the IQ modulator if iIN becomes to great.
To prevent the voltage drop in the input node, the common mode voltage for the OP also has to follow the envelope of the input signal. This will cause the common
Time A m p li tu d e (V ) envelope vinp vinn
Figure 4.1. Differential sinusoidal input signal and envelope tracking signal
Figure 4.2. Theoretical envelope tracking amplifier circuit
mode level of the OP output to decrease in proportion to the envelope signal and hence result in a lower current iN. So by varying iN and iCS in proportion to
the envelope signal, iIN can be maintained at the same level as when driving the
circuit without the envelope tracking current sources.
As the amplifier stage works as a current mirror for the mixer, the lowered bias current in the amplifier will also affect the mixer stage where the real benefit is achieved, because of the 70 copies of the mixer.
4.2
Implementation
The implementation of the Envelope Tracking IQM is rather simple, assuming that all components can handle the variations of the common mode voltage for
4.2 Implementation 39
Figure 4.3. Envelope Tracking IQM amplifier architecture
the chosen baseband frequency. By replacing the constant common mode voltage (CM) for the amplifier circuit with the envelope tracking signal (ET ), there will be a common mode mirroring of the currents through the N2 and N3 transistors, which will vary according to the amplitude of the input signal. The PMOS current mirrors at the top of the circuit will follow the same pattern and thereby act as adjustable current sources. The envelope signal was created by implementing an FWR (Full-Wave Rectifier) which is described in Section 4.2.1.
The RC filter of the circuit was also changed due to the high frequency com-ponents of the envelope tracking signal, this is described in Section 4.2.2. An overview of the Envelope Tracking IQM can be seen in Figure 4.4.
Name Value Rin 2 kΩ R 500 Ω Cc 411 fF
Cd 3.66 pF
Table 4.1. Component values of Envelope Tracking IQM
4.2.1
Rectifier
The design of the rectifier is far from optimized and was mainly constructed to give a hint on how to implement an FWR as an analog circuit and to show a rough example of the overhead current consumption of such a circuit. If implemented, it is most likely that the envelope signal will be designed as a digital signal, to give
Figure 4.4. Overview of Envelope Tracking IQM
the user full control over the shape and levels of the signal. Therefore, the analog FWR will not be described in detail, only a brief description of the functionality and the design will be presented here. The created circuit has an average current consumption of 705 µA and can be seen in figure 4.5.
Figure 4.5. Schematic view of the full-wave rectifier
The first block of the circuit creates a rectified shape of the input signal, referred to as the envelope signal. Block 2 subtracts the common mode voltage from this envelope to bring its minimum voltage level down to zero. To get a correct subtraction, the transistor sizes of block 1 and 3 are matched. The output of block 2 controls the current mirror in block 4, where the sizes of the PMOS transistors can be changed to adjust the amplification of the output. Block 5 is designed to match the transistor schematic of the amplifier (see figure 4.2) and the out signal
4.2 Implementation 41 is what will be replacing the common mode voltage.
An ideal FWR was also designed, using Verilog-A code and block 5 from Figure 4.5 to create a correct current mirror with the amplifier circuit. This is the FWR used for all simulations and is described by Figure 4.6. The Verilog-A block gen-erates an envelope tracking current with a fully adjustable swing with a maximum output of 153 µA, resulting in an output voltage of 960 mV. In the code example below the envelope tracking current has a full swing of 153 µA.
Figure 4.6. Ideal full-wave rectifier
// V e r i l o g −A f o r FWR
// G e n e r a t e s an e n v e l o p e t r a c k i n g c u r r e n t o f t h e d i f f e r e n t i a l i n p u t ‘ i n c l u d e " d i s c i p l i n e s . vams"
module ExFWR( inp , inn , out ) ; i n o u t inp , inn , out ; e l e c t r i c a l inp , inn , out ; r e a l vinp , vinn , i o u t ;
parameter r e a l vinmax = 300m; //Maximum i n p u t ac a m p l i t u d e parameter r e a l voutmin = 153u ; //Maximum o u t p u t c u r r e n t parameter r e a l voutmin = 0 ; //Minimum o u t p u t c u r r e n t a n a l o g b e g i n
vin p = V( i n p ) ; vin n = V( i n n ) ;
i o u t = ( abs ( vin p − vinn ) / 2 ) / vinmax ; // Normalized a b s o l u t e v a l u e i o u t = −( i o u t ∗ ( ioutmax−ioutmin )+ ioutmin ) ;
I ( out ) <+ i o u t ; end
4.2.2
RC filter
When implementing the envelope tracking into the circuit, a modification of the RC filter between the amplifier and mixer had to be made. The envelope signal has a lot of high frequency components (see Figure 4.7) which are suppressed by the present common mode filter of the Reference IQM with a cutoff frequency at 40 MHz. As described in Section 3.2.4, the purpose of the RC filter is to reduce the noise of the differential signal. This means that the cutoff frequency of the differential filter should be kept at the same value, whilst a wider bandwidth is wanted for the common mode filter. Observing the figure, one can see that the presented envelope signal has many high frequency components. One should be able to cut off a significant part of the highest frequencies without loosing too much performance. However, this thesis uses fully ideal envelope tracking.
The presented frequency spectrum is for the 10 MHz sinusoidal input used throughout the thesis. Though, the frequency spectrum will have different shapes for different signals, i.e. GSM, EDGE, WCDMA and LTE.
Figure 4.7. Frequency spectrum of envelope signal for 10 MHz sinusoidal input
By inserting a capacitor between the differential inputs of the mixer, the differ-ential filter gains an extra RC component of 2RCdwhich can be derived from the
green line in Figure 4.8. The resulting differential filter has the cutoff frequency presented by equation 4.2. The cutoff frequency of the common mode filter is found in equation 4.3 and is derived using the same relationship as in Section 3.2.4 (the red line).
f3dB_diff = 1
2πR(Ct+ 2Cd)
4.2 Implementation 43
f3dB_cm= 1
2πRCt
(4.3) Where Ct= Cc+ Cp.
Figure 4.8. Modified RC filter circuit for Envelope Tracking IQM
To widen the bandwidth of the common mode filter Ct has to be decreased.
This change will also increase the differential bandwidth if Cd = 0. That is why
Cd has to be increased when Ct is decreased to maintain the differential cutoff
frequency. The common mode filter cutoff frequency is set to 500 MHz for the Envelope Tracking IQM and with a maintained differential cutoff frequency of 40 MHz the resulting component values are; R = 500 Ω, Cc = 411 fF, Cd= 3.66 pF
and Cp still at 226 fF.
A potential problem with the different filter bandwidths, which has not been analyzed during this thesis work due to lack of time, is that different bandwidths results in different group delays. Meaning that the common mode part of the signal will propagate faster than the differential part because of the higher common mode bandwidth, causing an offset between the signal parts when arriving at the input to the mixer. An idea to solve this problem is to insert a delay for the envelope signal equal to the time offset of the common mode and differential signal parts.
(a) Differential filter
(b) Common mode filter
