Advanced silicon MMICs for mm wave automotive radar front ends

(1)

Advanced Silicon MMICs for

mm-Wave Automotive Radar

Front-Ends

vorgelegt von Master of Science Alexander Kravets geb. in Kiew, Ukraine

von der Fakult¨at IV – Elektrotechnik und Informatik der Technischen Universit¨at Berlin

zur Erlangung des akademischen Grades Doktor der Ingenieurwissenschaften

Dr.-Ing.

genehmigte Dissertation

Promotionsausschuss:

Vorsitzender: Prof. Dr. G. Tr¨ankle

Berichter 1: Prof. Dr.-Ing. habil. W. Heinrich Berichter 2: Prof. Dr.-Ing. Dr.-Ing. habil. R. Weigel Berichter 3: Prof. Dr.-Ing. N. Pohl

Tag der wissenschaftlichen Aussprache: 29. Oktober 2014

Berlin, 2015 D 83

(2)

(3)

Acknowledgement

I would like to express my gratitude to Prof. Wolfgang Heinrich and to Prof. Günther Tränkle for having offered me the opportunity to carry out this work at the Ferdinand-Braun-Institut, Leibniz-Institut für Höchstfrequenztechnik (FBH). Prof. Heinrich’s supervision, involvement and the fruitful discussions have been of help during the course of this work.

I would like to thank Prof. Robert Weigel and Prof. Nils Pohl for kindly accepting reviewing this thesis.

The FBH team deserve a big mention: I thank Dr. Udo Pursche for the fruitful cooperation and for the translation of the abstract to German, Prof. Dr. Matthias Rudolph for all things modeling, Dr. Franz-Josef Schm¨uckle for his EM wisdom and great mood, Ralf D¨orner and Jens Schmidt for their assistance with the measurements. I thank Dr. Chafik Meliani and Dr. Eldad Bahat-Treidel for the friendship and the support.

I thank Dr. Vadim Isaakov for his valuable advices.

I would like to thank Dr. Volker M¨uhlhaus for his support in usage of an EM simulator. My family’s contributions are too many to be succinctly summarized. Thank you.

(4)

Abstract

This work presents a high-linearity automotive radar front-end at 77 GHz in 0.25 µm SiGe technology. The passive elements are realized using thin-film microstrip lines. A detailed transformer balun synthesis procedure was developed. The realized passive baluns (“rat-race” and transformer) show excellent common-mode rejection ratios exceeding 30 dB and low losses of 2.5 dB. On the active side, a low-gain, high-linearity single stage common-emitter LNA was realized. The selected topology allowed finer trade-off between linearity and sensitivity of the front-end compared to multi-stage LNA solutions. For the mixer, a low voltage supply, high-linearity, low-noise double-balanced concept was employed. It uses AC-coupling between the two stages, which allowed an independent optimization of transconductance, core sizing and bias: the transconductance was designed for best noise performance, while the core was chosen for maximum linearity. A high-fidelity two-channel receiver was realized using these circuit components, which achieved a performance comparable to the published state-of-the-art results in SiGe: Single sideband noise figure better than 16.5 dB, 1-dB compression point exceeding -12 dBm, while consuming moderate 82 mA DC current from a 1.6 V supply for both channels.

(5)

Zusammenfassung

Mit der vorliegenden Arbeit wird ein besonders lineares 77-GHz-Front-end in 0,25 µm-SiGe-Technologie für Radaranwendungen in Fahrzeugen vorgestellt. Die passiven Komponenten wurden als Dünnschicht-Mikrostreifenleitungen hergestellt. Die so realisierten Baluns (rat-race-Koppler und Transformator) weisen eine hervorragende Gleichtaktunterdrückung (besser als 30 dB) und niedrige Verluste (ca. 2,5 dB) auf. An aktiven Komponenten wurde zunächst ein einstufiger LNA in Emitterschaltung entwickelt, der zwar eine geringe Verstärkung, dafür aber eine hohe Linearität aufweist. Im Gegensatz zu einer mehrstufigen LNA-Lösung stellt er einen besseren Kompromiss zwischen der Linearität des gesamten Front-ends und dessen Empfindlichkeit dar. Der Mischer wurde rauscharm und besonders linear nach dem doppelt balancierten Konzept mit niedriger Speisespannung realisiert. Durch Verwendung von Wech-selspannungskopplung zwischen beiden Stufen des Mischers konnten die Transistorgrößen und die Arbeitspunkte der beiden Stufen des Mischers getrennt optimiert werden: Die erste Trans-konduktanzstufe ist rauschoptimiert, die zweite Stufe, der eigentliche Mischerkern, ist für hohe Linearität ausgelegt. Aus all diesen Komponenten wurde ein Zweikanal-Empfänger aufgebaut, der den gegenwärtigen publizierten Stand der Technik in SiGe-Technologie repräsentiert: Einer einseitenbandbezogenen Rauschzahl von weniger als 16,5 dB und einem eingangsbezogenen 1-dB-Kompessionspunkt von -12 dBm stehen dabei eine Stromaufnahme von 82 mA aus einer 1,6-V-Speisespannungsquelle für beide Empfangskanäle gegenüber.

(6)

(7)

1 Introduction

1.1 Road Safety

Road safety is a major societal issue. While the likelihood of having a road crash for a single individual, on average, is in the range of once every 25 years (in developed countries), society as a whole pays a crushing price for the cumulative effects of crashes [13]. Annually worldwide between 20 to 50 million people suffer non-fatal injuries (with many incurring a disability) and 1.24 million people are killed because of accidents involving motorized vehicles. Road traffic injuries are the leading cause of death among young people, aged 15–29 years. On the roads of the European Union in 2012 28,000 people died, i.e. the equivalent of a medium town, and about 250,000 people were seriously injured [17].

In July 2010 The European Commission has adopted challenging plans to reduce the number of road deaths on Europe’s roads to one half by 2020 [16]. There are seven strategic objectives for this end:

1. Improved safety measures for trucks and cars 2. Building safer roads

3. Developing safer vehicles

4. Strengthening licensing and training 5. Better enforcement

6. Targeting injuries

7. A new focus on motorcyclists

Vehicle safety can be loosely separated into 2 categories: passive and active. Passive measures should reduce injury rate when an accident is inevitable. In this category are: seat belts, airbags, anti-lock braking systems (ABS) and electronic stability control (ESC). Active safety means having the ability to predict a situation which could probably result in an accident and having the capability to react to it timely so the accident could be avoided, or at least its’ impact is reduced (pre-crash sensing).

Human limitations in sensing and control multiply when hundreds of vehicles share the same roads at the same time. Traffic flow consisting of cars controlled by people is doomed to inefficiency due to human aspects of delayed response to traffic conditions. When we detect brake lights ahead, time is expended as we assess the situation and proceed to apply our own brakes, if needed. When traffic ahead accelerates, a similar lag time is incurred to sense that

(12)

Introduction

condition and follow suit. The aggregate effect of these factors creates “accordion effects” or “shock waves” in dense traffic flows, as well as the relatively slow clearance time for intersections controlled by traffic signals. Traffic congestion is also caused by the sheer volume of vehicles attempting to use roadways, exceeding physical capacity limitations.

Moreover, reducing the time interval from the recognition of the situation to the vehicle’s stop could greatly reduce collision probability. If the reaction time could have been improved by 0.25 s collision probability in case of rear-end collisions reduces by 30 % [35], [26]. Collisions at intersections could have been avoided in 50 % of all cases if the driver knew about the situation 0.5 s in advance. If it is too late to avoid the accident, the information about the location and the severity of the impact lets the vehicle take safety measures such that the risk of injuries is minimized. For example, the detection of an unavoidable side impact gives enough time to inflate airbags [21].

Planning engineers envision a cooperative ecosystem where vehicles exchange data with other vehicles (V2V), between vehicles and the infrastructure (V2I) and between infrastructures (I2I) in real time to have drivers better informed, reducing risks of accidents and making traffic flows run more efficiently and smoothly overall. The human inefficiencies noted above will be gradually moved to the domain of machine sensing and control. From a legal standpoint, sensors and computers are not allowed to make important decisions instead of humans on the highways, but to complement human natural sensors that were not suited for operation while moving at high speeds.

To create such a cooperative ecosystem:

a system based on fusion of sensors has to be made available

the costs of these technical solutions have to be brought down so they can be fitted onto mid- to lower-end of the vehicles

1.1.1 Automotive Comfort and Safety Systems

The term “comfort systems” (or “convenience systems”) came into being in the late 90ies when automotive companies were ready to offer driver-assist systems to their customers but were not yet ready to take on the legal implications and performance requirements that would come with introducing a new product labeled as “safety system” [13].

Comfort Systems Overview

Parking Assist. The simplest form of such system is a rear-facing camera, which simply offers a view of the area behind the vehicle but no driver warnings. Warning capability can be added by an ultrasonic sensor, covering the immediate area around the car. More advanced systems use short range radar to cover an extended range and provide the driver with more precise information as to the location of any obstacle.

(13)

Road Safety

Adaptive Cruise Control (ACC). It is the primary comfort system for highway driving. ACC allows a driver to set a desired speed which is automatically reattained when the road ahead is unobstructed. If a close vehicle in front moves at a slower speed then throttling and brake of the host vehicle is controlled to match this of the slower vehicle. In the future ACC systems will be extended to be operational at low speeds including full-stop capability (also known as “stop-and-go ACC”). Systems currently on the market monitor the scene using either long-range-radar or lidar (laser radar), future systems may also use machine vision.

Lane-Keeping Assistance (LKA). It uses computer vision technology to detect the lane in which the vehicle is traveling and adds torque to the steering wheel in order that the vehicle stays on the lane.

Safety Systems Overview

Main safety systems can be classified into the following sub-groups [69], [13]: Driver Perception Assistance

– Adaptive headlights: vehicle speed is taken into account to control illumination pattern.

– Night vision: helps to see objects far behind the view of the vehicle lights; is realised either with passive infrared camera or long range radar.

Crash Prevention

– Forward collision warning / mitigation / avoidance. Some European car manufac-turers implement only ACC (marketed as a comfort feature), while some Japanese manufacturers added an active brake assist for collision mitigation. Unlike the smooth deceleration of ACC system, the active brake assist provides much higher forces for deceleration. Thus comfort features of ACC are extended by the active brake assist into the safety domain. Systems to prevent forward collisions rely on long range radar or lidar sensing, sometimes augmented by machine vision.

– Lane departure warning and line change support. Machine vision techniques are used to monitor the lateral position of the vehicle within its lane.

– Side object warning (or “blind spot monitoring” ). A short range radar is used to monitor the rear left/right sectors obscured by the car’s carcass.

– Pedestrian recognition and warning. Sensing systems based on machine vision per-form real-time processing to detect the pedestrians and asses the potential danger. Pre-crash Sensing. Radar data can be combined with ABS data so if an imminent accident is identified by sensors and/or car dynamics then the brakes are pre-charged, the seat belt are pretensioned, airbags pre-fired, seat orientations will be adapted, sunroof will be closed, etc.

(14)

Introduction

The aforementioned sensor technologies, such as infrared camera, video camera, ultrasound, lidar have each their own specialized advantages [56], [69]. However, radar is the only sensor that:

can measure both range and angle-of-arrival accurately of both moving and stationary objects

is capable of directly providing velocity information

is not affected by weather conditions: it retains its performance in fog/rain/snow can be mounted invisibly behind plastic fascia

This suggests radar as the most robust technology for a vehicular environment. Additionally, radar can simultaneously integrate several automotive comfort and safety features.

1.2 Automotive Radar Applications – Brief History and Status

First experiments with automotive radar took place in the late 50ies [91]. Cadillac Cyclone (Fig. 1.1), General Motors’ concept car from 1959, was a first car to feature a radar. Behind the shiny black plastic cones (radomes) a modified airplane radar was installed. A proximity warning device was supposed to prevent collisions but it was never tested in practice.

Figure 1.1: Cadillac Cyclone, General Motors, 1959

In the 70ies the development moved to microwave frequencies. From the early beginning the key driver was collision avoidance. Due to progress both in semiconductor microwave sources, such as Gunn diodes and GaAs MMICs and to quickly evolving micro-controllers and DSPs the commercialization of automotive radar became feasible in the 90ies. in 1993 Greyhound (USA) installed more than 1600 Radar collision warning systems in their bus lines [36] yielding a reduction of accidents of 21 % (compared to the year before). The 24.125 GHz pulse-Doppler

(15)

Automotive Radar Applications – Brief History and Status

radars were developed by Vorad Safety Systems, San Diego and deployed a flat six-by-eight inch phased-array antenna.

In 1999 Daimler introduced 1st commercial radar ACC “Distronic” [109], which operated at 76.5 GHz and used the pulse-Doppler principle. Other automotive manufacturers followed with equipping their top-models with radars.

1.2.1 Automotive Short-, Mid- and Long-Range Radar

Automotive radar is classified according to covered ranges and azimuths (table 1.1). For Short Range Radar (SRR), ranging accuracy and large field of view are more critical than for Long Range Radar (LRR). These are addressed by higher bandwidth, multiple radar placement and antenna design. As explained in chapter 2 the dynamic range of the receiver is very critical in LRR case.

Table 1.1: Classification of Automotive Short-, Mid- and Long-Range Radar [67], [56]

Detection Range Ranging Accuracy Field of View Bandwidth [m] [m] [◦] [GHz] SRR 0 − 30 0.05 90 − 180 4− 5 MRR 2 − 150 0.1 30 − 60 0.5− 1 LRR 20 _{− 250} 0.5 5 _{− 20} 0.2_{− 1}

1.2.2 Automotive Radar Frequency Band Regulation

25 − 29 GHz: band allocated in North America for automotive ultra-wide band (UWB) SRR systems

22 − 26.65 GHz: allocated by EC (ETSI EN 302 288-1) to be deployed from 2005 till 2013 with penetration rate restricted to 7 % of all cars in each EC country (so that the other services in vicinity of 24 GHz)

24.05 − 24.25 GHz: license-free ISM band, can be used for CW radar (since it does not have enough bandwidth for pulsed radar). Since lane change assistance does not require highest range resolution, this band is deployed for MRR applications

76 − 77 GHz: this band is standardized in Europe (ETSI EN 301 091) for LRR, is being allocated for Intelligent Transport Services (ITS) in Europe, North America, Japan 77 − 81 GHz: EU, Japan, North America allocated this band for UWB SRR and MRR.

In EU this band usage was permitted from 2005 onwards

(16)

Introduction

1.2.3 Multi-Channel Automotive Radar Front Ends

Multi-channel front ends are of great interest for current and future applications in vehicu-lar and other fields. The receive channel diversity could be deployed to better exploit the information contained in the received signals for the following purposes:

Angle-of-arrival estimation is required in LRR. Among other techniques this can be achieved by antenna- and receiver-diversity [95] with conjunction with parameter esti-mation methods, based on subspace techniques [31], [63].

The information available from several receive channels could be deployed as an electronic beam former (also known as digital beam former), by steering the beam away from an interferer towards the desired signal. Therefore the immunity to multi-path scattering present in real-world conditions is increased [72], [109]. The electronic beam former is an alternative to mechanically beam steering, which is sensitive in mechanical reliability over lifetime and is limited with regard to miniaturization [91].

Coherent summation of the multi-receiver outputs improves signal-to-noise ratios thereby improving the equivalent receiver sensitivity.

1.3 Thesis Objectives and Organization

The goal of this dissertation is to present circuit blocks to implement a 77 GHz multi-channel receiver for automotive radar using SiGe technology. The emphasis is put on high linearity and sensitivity of the receiver, for a very high dynamic range is crucial for such instrumental applications.

Beside this:

development and exploration of alternative design methodologies suitable for designs at the technology margins (fmax/3)

minimizing the power consumption by reducing both the circuit blocks’ supply voltages and quiescent currents and selecting power-efficient circuit topologies

integration – reducing of the number of separate chips The thesis is organized as follows:

In chapter 2 automotive FMCW LRR performance is assessed on system level. Two cases of FMCW radar transceiver combining (quasi-bi-static and mono-static) are compared in presence of several leakages and reflections.

Chapter 3 discusses the circuit environment. Additionally, models are developed for the pas-sivated microstrip transmission line model and for the prober pads.

(17)

Thesis Objectives and Organization

Chapter 4 focuses on design of active and passive circuits, such as LNA, mixer, several balun types. Various topologies and circuit- and system-level design trade-offs (especially linearity vs. noise) were juxtaposed. The measured circuits are benchmarked with the published results. Chapter 5 deals with the integration of the circuits from the previous chapter into 1- and 2-channel receivers. For the 2-2-channel receiver an active LO splitter is designed and isolation improvement techniques are discussed. The measured receivers are benchmarked with the published results.

Chapter 6 wraps up this work with summary, contributions’ outline and possible future inves-tigations’ suggestions.

(18)

(19)

2 Automotive LRR FMCW Radar – System

Level Approach

2.1 FMCW Radar Fundamentals

2.1.1 Waveform Derivation

Frequency Modulated Continuous Wave Radar is a technique for obtaining range information. FMCW Radar has a long history [64], in the past its use was limited to specialized applications, such as radio altimeters. Linear frequency modulation is very versatile when applied to an optimal receiver (also known as correlation receiver) [79] and exists in nature: big brown bat deploys dual linear frequency modulated ultrasonic radar to navigate and forage [48]. In a correlation receiver the transmitted signal is mixed with a delayed replica of itself – this is also known as homodyne receiver. A frequency measurement must be performed to obtain range from FMCW receiver. This is done in the digital domain using FFT (Fast Fourier Transform). The main advantage of FMCW radar is its high time bandwidth product (Tsw∆F ) [23], [41], [94]. High sweep time values improve the overall sensitivity (noise filter

bank bandwidth is inverse proportional to sweep time) while the range resolution is inverse proportional to sweep bandwidth.

Frequency modulation can be achieved in several ways. Here one the most common ways is considered – sawtooth modulation of the carrier.

The instantaneous frequency of the transmitted signal is (Fig. 2.1a): fTX= f1+∆F t

Tsw

[Hz] (0≤ t < Tsw) (2.1)

where f1 is the start frequency (76.5 GHz, in this case), ∆F – the sweep bandwidth of the

radar (1 GHz) and Tsw is the sweep time duration (typically in the order of magnitude of 1 ms).

Therefore, the instantaneous phase of the transmitted signal is:

ϕTX(t) = 2π t Z 0 fTX(t)dt = 2πf1t + 2π ∆F t2 2Tsw + ϕT [rad] (0≤ t < Tsw) (2.2)

where ϕT is the phase offset due to the integration. The instantaneous phase of the received

(20)

Automotive LRR FMCW Radar – System Level Approach signal is: ϕRX(t) = ϕTX(t− τ) = 2π t−τ Z 0 fTX(t)dt = (2.3) = 2πf1(t− τ) + 2π ∆F (t− τ)2 2Tsw + ϕT+ ϕR [rad] (τ ≤ t < τ + Tsw)

where ϕR is the phase offset added by the target and

τ = 2d c0

[s] (2.4)

is the delay of the signal occurring due to two-way propagation (the target is distanced d meters from the radar). As it is apparent from equation 2.4 and previous considerations,

τ _Tsw (2.5)

After mixing and low-pass filtering the Intermediate Frequency (IF) signal is obtained: ϕIF(t) =|ϕRX− ϕTX| = 2π∆F_T_swτ t + 2πf1τ − ϕR− π ∆F τ2 Tsw [rad] (2.6)

In equation 2.6, the most important term is the first, since it represents the beat frequency, proportional to the range. The second and third terms in equation 2.6 represent constant phase offset added by target and the last, fourth term can be neglected, since equation 2.5 holds.

fbeat, static= wbeat, static 2π = ∆F Tsw τ = ∆F Tsw 2d c0

[Hz] (where d is the unknown range) (2.7) Therefore, equation 2.7 yields fbeat, static= 33. . . 1333 kHz for d = 5. . . 100 m.

The beat signal spectrum is:

Sbeat, static(f ) = Tsw

Z

0

cos(wbeatt)e−jwtdt = (2.8)

= sin((wbeat− w)

Tsw 2 )

wbeat− w · e

jwbeat−w₂ Tsw₊ sin((wbeat+ w)Tsw₂ )

wbeat+ w · e

−jwbeat+w2 Tsw

It must be added that since in time domain the signal is periodic with period Tsw, its’ Fourier

transform is sampled with frequency 1/Tsw.

This derivation was performed for static targets. For targets moving towards the radar with radial velocity vr [95]:

(21)

FMCW Radar Fundamentals where: fDoppler= 2 vr c0 fTX [Hz] (2.10)

Equation 2.10 is known as the Doppler effect: for an observer moving relative to the wave source the frequency is changed.

Therefore the beat frequency contains a term proportional to the target’s distance and a term proportional to target’s velocity. In order to estimate both targets’ properties two sweeps (up- and down-, Fig. 2.1b) can be performed so 2 equations with 2 unknowns are formed. In scenarios where a radar can be jammed by neighboring radar, especially such as SRR the

RX

sweep time [s]

RF [Hz]

TX

∆

F

T

_sw

τ

(a) TX and Corresponding RX Frequencies

sweep time [s]

IF [Hz]

T

_sw

f

_r

+f

_doppler

τ

f

_r

f

_r

-f

_doppler (b) Corresponding IF

Figure 2.1: FMCW: Frequencies Plane vs. Time for positive Doppler Target Frequency CW radar can be modulated with predefined pseudo-noise sequences [21], [99], so the radar is “blind” to other pseudo-noise signatures as well as clutter.

2.1.2 Two-way Radar Equation

The transmitted power is reflected by the target and returns to the receiver (it is assumed that the transmitter and receiver are co-located, Fig. 2.2a and Fig. 2.2b). The received power can

(22)

Automotive LRR FMCW Radar – System Level Approach

be described by the following equation [94]: PRX= PTXGTXGRX σc0

2

(4π)3_f2_d2n [W] (2.11)

In the logarithmic form the two-way radar equation is: PRX= PTX+ 2G + RCS + 20 log(c0)− 10 log (4π)3f2

− 20n log(d) [dBm] (2.12) where:

PRX= +15 dBm

G = GTX= GRX is the TX and RX antenna gain, assumed to be 20 dB

RCS = 10 log σ stands for “Radar Cross Section”. Similar to the receiving antenna, a radar target also intercepts a portion of the power, but reflects it in the direction of the radar. The amount of power reflected back toward the radar is determined by RCS of the target. RCS is a characteristic of the target that represents its size as seen by the radar. In the linear scale radar cross section σ has the dimensions of area. Additionally, it has a strong dependence on the angle of incidence. For modern sedan cars: RCS = +3. . . + 17 dBsm [89]. In the following calculations RCS = +3 dBsm is assumed

n is the path loss exponent: for multi-path n = 2.5. . . 3; here LOS (line of sight) assumed n = 2

Table 2.1: Relationship between PRX and Corresponding Target Distance

PRX [dBm] d [m]

−51 5

−115 200

2.2 Leakages and Reflections in FMCW Radar

In pulse radar a narrow pulse in time domain is transmitted and the range is estimated from the time difference between the received (echo) signal and the transmitted signal (reference time stamp). Therefore, the time instances of transmission and reception don’t coincide, so the RX is inherently isolated from the TX. In contrast to pulse radar, FMCW radar’s major drawback is linked to its correlative nature: simultaneously receive while transmitting so there is an inherent RX sensitivity degradation due to TX leakage into the RX. Duplexers widely used to separate TX and RX communication bands cannot be used since TX and RX frequencies are essentially the same in FMCW radar. In this application (LRR) the transmitted signal power is +15 dBm while the received signal from the furthest target is −115 dBm. Even

(23)

Leakages and Reflections in FMCW Radar

were the overall TX to RX isolation 40 dB (an optimistic estimation, regardless of antenna arrangement) then the TX leakage signal is 90 dB above the desired RX signal. This would mean, that ideally the front end has to accommodate such dynamics and still remain linear enough. Otherwise, false targets’ indications due to spectral regrowth will appear. This strict linearity requirement, simultaneously in conjunction with low noise figure specification (needed to cover the maximal range) would require a very challenging front end. It also may explain why pulse radar approach is historically more common than FMCW – due to inherent mutually-exclusive TX-RX time-domain multiplexing.

2.2.1 Automotive Radar Classification According to TX-RX Separation, System Architecture and Trends

TX

RX

(a) Quasi-Bi-static

TX

/

RX

(b) Mono-static

Figure 2.2: Quasi-bi- vs. Mono- Static Radar

when separate TX and RX antennas are deployed the arrangement is called quasi-bi-static [36] (Fig. 2.2a). Quasi-bi-static radar arrangement avoids the most severe mono-static radar leakages and reflections (Fig. 2.5).

The main drawback of quasi-bi-static arrangement is the space consumed by an addi-tional antenna, which could be limiting factor in several automotive radar applications. Hence, a compromise where TX and RX share same antenna, called mono-static radar (Fig. 2.2b, 2.4). It is an attractive solution due to its compactness, despite leakages and reflections (Fig. 2.6).

A multi-static radar system is one in which there are at least three components - for example, one receiver and two transmitters, or two receivers and one transmitter, or multiple receivers and multiple transmitters. By definition, it is a generalization of the quasi-bi-static radar system, with one or more receivers processing returns from one or more separated transmitters. This is the case in automotive LRR applications, where if angle measurement are crucial

(24)

multiple receivers with multiple antennas are deployed [95], [91]. Additionally, both at RX and TX sides antenna count might be increased for digital beam-forming purposes.

In terms of commercial automotive radar systems there is a continuum of solutions [67]: from low-end single antenna mono-static radars ICs (1× RXN7740) to high-end single TX antenna, 16 RX antennas (1 _{× RTN7730, 4 × RRN7740). Since the early 2000 the trend for the} high-end automotive radar sensors is in favor of multi-channel architecture, driven by the digital beam-forming applications and advances in array signal processing theory [91], [72], [109].

2.2.2 Antenna-Transceiver Combining

Fig. 2.3 and Fig. 2.4 show the antenna-transceiver interfaces for the quasi-bi-static and mono-static cases. The quasi-bi-mono-static case (Fig. 2.3) is straightforward. The mono-mono-static interfaces can be further subdivided in 2 subgroups:

Circulator (Fig. 2.4a): conceptually a circulator (a passive ferromagnetic non-reciprocal three port [15]) lends itself perfectly for the combiner: power entering any port is trans-mitted to the next port in rotation, with the third port being isolated. The drawbacks normally associated with circulators at such frequency are: performance (isolation, in-sertion loss), availability, size, integration (it cannot be integrated on the IC).

Coupler (Fig. 2.4b): a 3 dB branch-line coupler divides the input signal between 2 outputs, therefore theoretically it is a less attractive option. In the TX mode only half of TX signal reaches the antenna (the other half goes to the LO) and in the RX mode only half of received signal reaches the LNA (the other half is dissipated on the PA output impedance), therefore the theoretical SNR reduction (compared to the circulator case) is 6 dB. Despite this limitation branch-line couplers can be integrated on chip, with isolation up to 25 dB and insertion loss of circa 1 dB [43], therefore are preferred in current mono-static LRR implementations.

LO Source PA Front-End Coupler IFout LO Buffers LOin RFin RX Antenna TX Antenna

(25)

Leakages and Reflections in FMCW Radar LO Source PA Front-End Circulator _Antenna Coupler IFout LO Buffers LOin RFin (a) Circulator LO Source PA Front-End Antenna IFout LO Buffers Coupler LOin RFin (b) Coupler

Figure 2.4: TX-RX Combining Approaches: Mono-static

2.2.3 FMCW Radar with System Non-idealities

The system non-idealities, such as leakages and reflections are shown at Fig. 2.5 and Fig. 2.6 for quasi-bi static and the mono-static cases, respectively. The mixer LO-RF leakage is shared by both architectures. The quasi-bi-static case has only a single additional leakage source of the TX antenna into the RX. In the mono-static case an additional coupler / circulator leakage is added and also the reflections from all the discontinuities: Flip-Chip transition, antenna, radome, bumper. In order to make the discussion more quantitative, the scenario of leakages and reflections shown at table 2.2 is assumed. PLO is assumed to be +4 dBm and

PTX = +15 dBm. The quasi-bi-static non-idealities are of higher order as compared to the

mono-static case:

in the mono-static case the mixer isolation has a secondary effect on the overall system, since the power applied to mixer LO is an order of magnitude less than the TX power leaking through the coupler and reflected from 4 discontinuities

TX-RX antenna leakage doesn’t lend itself for easy modeling, since the leakage depends on the specific antenna and radome realization, antenna radiation curve and side-lobes suppression, antennas’ physical separation and the specific enclosure environment (the

(26)

reflections from bumper are also hard to model since they cannot be anymore approxi-mated by a single reflection orthogonal to bumper plane).

Table 2.2: Fig. 2.6 Blockers’ Parameters

Contributors [dB] Delay [mm] IF [Hz]

Leakages (Isolation) Mixer (LO to RF) -30 0.1 0.67

Coupler / Isolator -30 0.3 2

Reflections (Return Loss)

Flip Chip -20 0.5 3.33

Antenna -20 10 66.67

Radome -20 50 333.33

Bumper -20 100 666.67

The blockers’ effects will be analyzed as applied to the mono-static case and the quasi-bi-static case (which could be approximated as a private case of mono-quasi-bi-static arrangement) will be inferred from it.

LO Source PA Front-End Coupler IFout LO Buffers LOin RFin Mixer LO-RF TX-RX TX Radome Bumper Leakage leak. Target refl. TX Antenna RX Antenna Antenna RX Radome

Figure 2.5: Leakages and Reflections in Quasi-Bi-static Automotive FMCW Radar

Noise Floor with Blockers Present

The blockers in conjunction with realistic phase noise profile of the TX VCO introduce a potential problem (Fig. 2.7). The leaking/reflected TX phase noise “skirt” is introduced, obscuring the targets’ return signatures which are further away and/or having small RCS or, at least, reducing the SNR.

Range Correlation Effect – IF Phase Noise Improvement In a presence of a blocker the equivalent transfer function H(f ) translates the transmitted phase noise spectral density to

(27)

Leakages and Reflections in FMCW Radar LO Source PA Front-End Circulator Coupler IFout LO Buffers LOin RFin Mixer LO-RF Circulator Flip-Chip Antenna Radome Radome Bumper refl. refl. refl. Bumper refl. leak. leak. Target refl.

Figure 2.6: Leakages and Reflections in Mono-static Automotive FMCW Radar

RX Signal

Noise Floor

IF [Hz]

Power [dBm]

Blocker

Figure 2.7: Target Masking by Blocker Phase Noise Profile IF: SIF(f ) = STX(f ) H(f)2 [W_/_Hz_] _(2.13) In logarithmic scale: SIF(f ) = STX(f ) + 10 log H(f)2 [dBm_/_Hz_] _(2.14)

STX(f ) and SIF(f ) are the transmitted and the resulting IF phase noise spectral densities,

respectively. For a single blocker, the transfer function reads:

H(f ) = 1− e−j2πfTdelay (2.15)

where d_{leakage / reflection path} is the physical length of the reflection path and the equivalent one-way length of the leakage path, respectively:

Tdelay =

2d_{leakage / reflection path} c0 [s] (2.16) H(f)2 = 2 1− cos(2πfTdelay) (2.17) 17

(28)

Substituting equation 2.17 in equation 2.14 yields SIF(f ) = STX(f ) + 10 log

2 1_{− cos(2πfT}delay)

[dBm_/_Hz_] _(2.18)

Keeping in mind that: sin2α 2 = 1 2 1_{− cos α} (2.19)

The approximation in equation 2.20 holds for small α values, which is the case here (only the first term of the Taylor series):

1_{− cos α}= 2 sin2α 2 ≈ 2 ·α 2 22 = α2 2 (2.20)

Further substitution of equation 2.20 in equation 2.18 yields

SIF(f )≈ STX(f ) + 20 log(2πf Tdelay) [dBm/Hz] (2.21)

The effect shown in equation 2.21 is called “Range Correlation”: for small 2πf Tdelaythere is a

significant suppression of TX phase noise sidebands at the IF. TX phase noise passes through a high pass filter with a slope of +20_decadedB . The amount of correlation (and thus noise filtering) depends upon the range of clutter being illuminated, i.e. the time delay between the transmitted and received signal. Specifically, the correlation and amount of filtering is large at short ranges and becomes smaller as the range increases [11]. This effect reduces the detrimental effect of target masking shown at Fig. 2.7 and enables homodyne FMCW detection using available signal sources in presence of front-end leakages and reflections. Amplitude noise is significantly lower than phase noise [12] (more than 25 dB below) and therefore can be neglected.

In order to numerically evaluate the resulting IF phase noise power density spectrum a TX phase noise power density spectrum is needed. Reference [45] report on free-running and phase-locked VCOs in 0.18 µm SiGe for 77 GHz automotive FMCW radar. The reported phase noise at 100 kHz offset is −77dBc

Hz

and −102dBc Hz

for the free-running and the locked cases, respectively.1 The phase-locked VCO phase noise, as expected, is better at frequencies below PLL loop bandwidth (_{≈ 1 MHz), where the superior phase noise of the reference crystal} dominates.

Both phase noise characteristics from [45] and the resulting IF phase noise spectrum from blockers (table 2.2, equation 2.21) are shown on Fig. 2.8 and 2.9. The free-running VCO phase noise has slope of ≈ −20 dB

decade

(Fig. 2.8), after being converted to IF the free-running phase noise profile flattens. In the phase-locked case the TX phase noise plateaus from 10 kHz until 1 MHz so at the IF (Fig. 2.9) the phase noise assumes the ≈ +20decadedB

slope. Both figures and equation 2.21 indicate that the TX phase noise is suppressed the most for the closest blockers, while even for the furthest blockers (1.33 MHz) the phase noise improvement is circa 45 dB.

1_{[87] mention}

−75dBc

Hz at 100 kHz offset in a phase-locked state, which is product-oriented design accounting also for temperature variation, process and voltage spread.

(29)

Leakages and Reflections in FMCW Radar 103 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −240 −200 −160 −120 −80 −40 0 frequency offset [Hz] Phase Noise h dBc Hz i TX IF (mixer LO - RF leakage) IF (coupler leakage) IF (Flip Chip transition reflection)

IF (antenna reflection) IF (radome reflection) IF (bumper reflection)

Figure 2.8: IF Phase Noise: Range Correlation Effect on the Free-Running TX

103 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −240 −200 −160 −120 −80 −40 frequency offset [Hz] Phase Noise h dBc Hz i TX IF (mixer LO - RF leakage) IF (coupler leakage) IF (Flip Chip transition reflection)

IF (antenna reflection) IF (radome reflection) IF (bumper reflection)

Figure 2.9: IF Phase Noise: Range Correlation Effect on the Phase Locked TX

Noise Floor Power Contributors’ Breakdown The thermal noise floor power at the IF reads (logarithmic notation is more convenient and thus is preferred here):

Pthermal noise(f ) =−174 + NFSSB+ 10 log(1/Tsw) [dBm] (2.22)

The last term in equation 2.22 is the equivalent noise bandwidth due to sweep time in FMCW radar. The longer is the sweep duration, the narrower is the noise bandwidth so less noise power enters the receiver. For Tsw = 1 ms and NFSSB= 16 dB one obtains thermal noise floor

(30)

of _{−128 dBm. Blockers’ powers:}

Pmixer leakage(f ) = PLO+ Isolationmixer+ SIF(f ) + 10 log(1/Tsw) [dBm] (2.23)

In all of the blocker equations the term SIF(f ) assumes appropriate Tdelay. Unlike the rest of

the blocker cases in the mixer isolation case (equation 2.23) PLO is used.

Pcoupler leakage(f ) = PTX+ Isolationcoupler+ SIF(f ) + 10 log(1/Tsw) [dBm] (2.24)

Prefl(f ) = PTX+ RL + SIF(f ) + 10 log(1/Tsw) [dBm] (2.25)

Equation 2.25 for the reflected scenario is applied to 4 cases: Flip-Chip, Antenna, Radome and Bumper (RL stands for return loss).

Ideally, for the blockers not to affect the sensitivity of the front-end (limited by its NFSSB)

each of the blockers has to be significantly below the thermal noise floor (10 dB is taken as a safety margin):

P_{leakage / refl}(f )_{≤ P}thermal noise(f )− 10 [dBm] (2.26)

As it is evident from Fig. 2.10 and Fig. 2.11 the condition in equation 2.26 is satisfied for all blockers apart of radome and bumper reflections (in the phase locked case the blocker noise floor rises with the frequency at 1 MHz offset due to the typical PLL phase noise hump around the loop bandwidth). Therefore, in a mono-static arrangement the sensitivity is limited by the blockers (and not by the front-end noise figure, as usual), despite the range-correlation benefits. From practical point of view, Fig. 2.11 indicates that the mono-static system is still applicable for targets up to 200 m, but range accuracy reading for maximal ranges will be degraded (since it is a function of SNR), which is acceptable for distant targets. For targets closer than 100 m as Fig. 2.11 shows the thermal noise dominates.

Analysis of Blockers Passing through a Non-linear Transfer Function

As explained in section 2.2 the multi-tone excitation, consisting of several blockers and the target return is applied to a nonlinear front end, creating and enhancing the previously non-existing frequency components, both in and out of the IF band of interest (0.033 . . . 1.333 MHz). The same reflections/leakages scenario as outlined in subsection 2.2.3 is assumed, see Fig. 2.6. The 3rd _{order intermodulation products resulting from 2 blockers alone could be calculated}

using the common 2-tone Intermodulation Distortion (IMD) calculation [78], [55]:

iIPn = Pin+ IMD n_{− 1} [dBm] (2.27) iIP3 = Pin+ IMD 2 [dBm] (2.28)

(31)

Leakages and Reflections in FMCW Radar 103 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −200 −180 −160 −140 −120 −100 frequency offset [Hz] P ow er [dBm] Thermal Noise Mixer LO-RF Leakage Noise

Coupler Leakage Noise Flip-Chip Transition Reflection Noise

Antenna Reflection Noise Radome Reflection Noise Bumper Reflection Noise

RX Signal

Figure 2.10: IF Noise Floor Contributors – Free-Running TX

103 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −240 −220 −200 −180 −160 −140 −120 −100 frequency offset [Hz] P ow er [dBm] Thermal Noise Mixer LO-RF Leakage Noise

Coupler Leakage Noise Flip-Chip Transition Reflection Noise

Antenna Reflection Noise Radome Reflection Noise Bumper Reflection Noise

RX Signal

Figure 2.11: IF Noise Floor Contributors – Phase Locked TX

Thus, for an IP3 of 0 dBm and−5 dBm of input power one obtains IMD of 10 dB. This means that the resulting IMD products will be 10 dB below the carrier, which is the_{−15 dBm leakage} in this case, or 100 dB above the weakest target return of _{−115 dBm. The spectral regrowth} is harsh despite state-of-the-art linearity of the front-end deployed (iP1dB=−10 [dBm]).1

The main goals of this part are:

numerically investigate several multi-tone blocker and signal combination scenarios for

1_{In this text it is assumed iIP3}

≈ iP1dB + 10 [dBm], [78].

(32)

various front-end linearity values

assess the extent of applicability of either mono-static and/or quasi-bi-static architectures in presence of blockers

Non-linear Transfer Function Description with System-Level Parameters Large signal sim-ulation with a specific circuit-level implementation would yield results only at a certain linearity parameters’ set values and will be valid only for that circuit implementation. In order not to lose generality the large signal performance was investigated using a behavioral model. Two assumptions are made for front-end behavioral modeling:

frequency conversion and RF-path non-linearity can be treated separately

frequency conversion can be approximated as an ideal conversion from the carrier fre-quency to IF

(Both assumptions are valid for low-IF frequency converters, such as this case). Therefore the front-end performance under multi-tone excitation, its harmonics and intermodulations were simulated at IF frequency only [36] (as opposed to several combinations of RF and LO frequencies, their harmonics and intermodulations). This results in faster simulation times and lower memory requirements. A generic non-linearity was deployed to model the front-end by explicitly specifying front-end IP2 and IP3 values as shown in Fig. 2.12.1

P = PTX+RLRADOME=-5 dBm f = IFRADOME=333.33 Hz P = PRX_no2=-115 dBm f = IFRX_no2=1.333 MHz P_1Tone P_1Tone Ideal Passive Power Combiner 3 dB Amp Front-End TOI=iIP3 SOI=iIP2 G=1 NFSSB=16 dB IF

Figure 2.12: Front-End Linearity Case Study 1 Set-Up: A Weak RX Signal in Presence of a Single-Tone Jammer

In the small signal regime the front-end transfer function from RF to IF port can be approxi-mated by a simple polynomial of 3rd order:

y≈ a0+ a1x1+ a2x2+ a3x3 (2.29)

where x is the RF input voltage, y – the approximated IF output voltage, a0 – the DC output

voltage (circa 0.8 V), a1 = 1 is the voltage gain. Here a similar approach to noise 2-port

representation is taken: the front-end is represented as a cascade of non-linearity (with unity

(33)

gain) and of linear gain stage with a proper gain value. The latter, linear gain stage adds no information to this description, so is omitted; and the unity gain of the chosen representation would ease comparison of the multi-tone input to “jammed” output. Evaluating 2.29, one can show that [55]:

iIP2 = 20 loga1 a2

[dB] (2.30)

iIP3 = 20 logr4a1 3a3

[dB] (2.31)

Therefore, a normalized small-signal non-linearity can be approximated by 2 main linearity specifications: IP2 and IP3. Harmonic balance simulation engine re-calculates the polynomial coefficients as function of the input signal, since from certain drive level a1 value would

de-crease due to compression.1 Harmonic balance simulations were performed with the following settings:

Order: the blockers were simulated up to 3rd _{order (large non-linearities) and the signals}

were simulated up to 2nd order (weak non-linearities)

MaxOrder = 3: MaxOrder is the maximal order of the intermodulation terms assumed in the simulation. The value of 3 accounts for both IP2- and IP3- related effects (2ndand 3rd _{intermodulation orders, respectively). Higher MaxOrder values were not considered}

due to both exponential increase in simulation time and for avoiding unnecessary clutter in the results

A multi-dimensional sweep of various setup and system parameters was performed: 2 different targets/blockers scenarios were considered

1. Single 200 m target, single_{−5 dBm blocker} 2. Single 200 m target, two (−5 dBm each) blockers front-end iIP3

1. upper bound from realization point of view (iIP3= +20 dBm) 2. high-linearity front-end (iIP3= 0 dBm)([97], this work) 3. low-linearity front-end (iIP3=−20 dBm) ([46], [106], [3], [73]) front-end iIP2

1. upper bound from realization point of view (iIP2= +60 dBm) 2. good (iIP2= +40 dBm)

3. low (iIP2= +20 dBm)

1_{This happens due to the spectral components caused by a}

3x3at the signal frequency.

(34)

IP3 Sweep

Case Study 1: A Weak RX Signal in Presence of a Single-Tone Jammer Simulation setup is shown on Fig. 2.12: a single target at 200 m (1.333 MHz), single −5 dBm blocker (333 Hz)

Fig. 2.13a shows the full IF signal spectrum, from DC to 8th _{harmonic of the fundamental}

tone. The figure will be used to assess various front-ends’ responses to the blocker and its harmonics: 102 ₁₀3 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −160 −120 −80 −40 0 IF [Hz] P ow er [dBm] Blocker RX Signal iIP3 = +20 dBm iIP3 = 0 dBm iIP3 =−20 dBm

(a) 1 Target, 1 Blocker

102 ₁₀3 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −160 −120 −80 −40 0 IF [Hz] P ow er [dBm] Blockers RX Signal iIP3 = +20 dBm iIP3 = 0 dBm iIP3 =−20 dBm (b) 1 Target, 2 Blockers

Figure 2.13: Various Targets/Blockers Scenario: IP3 swept, iIP2 = +40 dBm

In theoretical case the fundamental is correctly registered; the higher harmonics are more than 50 dBc down

In the high-linearity case the fundamental is compressed by 3 dBc; the 2nd _{harmonic is}

49 dBc down (relative to fundamental) and the 3rd _{harmonic is 17 dBc down (relative to}

fundamental)

In the low-linearity case the fundamental is compressed by 22 dBc; the 2nd _{harmonic is}

89 dBc down (relative to fundamental), the 3rd harmonic is 10 dBc down (relative to fundamental)

The theoretical and high-linearity front-ends are coping adequately with the blocker alone, with the low-linearity front-end saturated already.

Since the blocker is between the DC and the IF band, it is possible to filter it out (and its harmonics) using a band pass filter at IF (either in HW or SW).1The intermodulation products of the weak signal (and its higher harmonics) with strong blocker fundamental (and its higher harmonics) are within the IF BW.

1_{The blocker and its harmonics were investigated in order to gain fuller understanding of energy spectral} regrowth.

(35)

Fig. 2.14a shows an overview of the target signal in the IF band for various front-ends:

1.33 1.33 1.34 1.34 1.34 ·106 −160 −140 −120 −100 −80 −60 −40 −20 0 20 IF [Hz] P ow er [dBm] Thermal Noise RX Signal iIP3 = +20 dBm iIP3 = 0 dBm iIP3 =−20 dBm

1.33 1.33 1.34 1.34 1.34 ·106 −160 −140 −120 −100 −80 −60 −40 −20 0 20 IF [Hz] P ow er [dBm] Thermal Noise RX Signal iIP3 = +20 dBm iIP3 = 0 dBm iIP3 =−20 dBm (b) 1 Target, 2 Blockers

Figure 2.14: Various Targets/Blockers Scenario: IP3 swept, iIP2 = +40 dBm; Zoomed to IF In the theoretical case: the signal is correctly registered (−115 dBm)

In the high-linearity case: the reading is enhanced by 39 dBc In the low-linearity case: the reading is enhanced by 66 dBc

The intermodulation between the blocker, its harmonics and the signal is the source for the spectral regrowth around the target. These intermodulations are present in the high linearity case and are quite strong in low linearity case. The theoretical front-end has a performance edge in this case (vs. high linearity front-end), while low linearity front-end has the worst performance.

Case Study 2: A Weak RX Signal in Presence of Two Jammers Simulation setup is shown in Fig. 2.15: a single target at 200 m (1.333 MHz), two _{−5 dBm blockers (at 333 Hz} and 667 Hz):

Fig. 2.13b shows an overview of the blocker and its harmonics: the situation is similar to blocker in Case Studies 1. The main differences are that more frequency components are present here due to intermodulations between the blockers and that the blockers here are slightly more compressed. Additionally, the intermodulations worsen due to doubling of blocker power. Fundamental blocker harmonic compression and intermodulations worsen with front-end linearity decrease.

Fig. 2.14b shows an overview of the signal, its harmonics and its intermodulations with the blockers in the IF band:

In the theoretical case: the dominating component in signal fundamental vicinity is 60 dBc up (relative to the original signal power); 2nd_{harmonic vicinity regrowth is below}

the noise floor

(36)

Automotive LRR FMCW Radar – System Level Approach P = PTX+RLRADOME=-5 dBm f = IFRADOME=333.33 Hz P_1Tone P_1Tone Ideal Passive Power Combiner 4.77 dB Amp Front-End TOI=iIP3 SOI=iIP2 G=1 NFSSB=16 dB IF P = PRX_no2=-115 dBm f = IFRX_no2=1.333 MHz P_1Tone P = PTX+RLBUMPER=-5 dBm f = IFBUMPER=666.67 Hz

Figure 2.15: Front-End Linearity Case Study3 Set-Up: Weak RX Signal in Presence of Two Jammers

In the high-linearity case: the dominant component in signal fundamental vicinity is 91 dBc up relative to the original fundamental power, 2nd _{harmonic vicinity regrowth is}

37 dBc down (relative to dominating component in fundamental vicinity)

In the low-linearity case: the dominant component in signal fundamental vicinity is 74 dBc up relative to the original fundamental power, 2nd _{harmonic vicinity regrowth is}

33 dBc down (relative to dominant component in fundamental vicinity)

While the blocker situation here resembles the one predicted in Case Study 1, the situation at higher frequencies is worse. Even the theoretical front-end is not sufficent – the intermodu-lations are 60 dBc higher than the signal (the high-linearity front-end yields intermoduintermodu-lations 90 dBc above the signal). Interestingly, the low-linearity front end seems to be more linear than the high-linearity front-end due to lower intermodulations around the signal. This is mis-leading, since due to both blockers most of the energy is spread around the DC and thus there is less power available in the vicinity of the signal. Additionally, the low-linearity front-end is so deep into saturation that the intermodulations cannot grow with an additional blocker anymore. As a result, the power is spread over a wider bandwidth (as compared to a single blocker case, Fig. 2.13a).

IP2 Sweep It is useful to have a measure indicating which of the two mechanisms (IP2 or IP3) limit the front-end performance in given blocking conditions [82]:

Pcorner blocker = 2iIP3− iIP2 [dBm] (2.32)

below Pcorner blockerthe front-end is IP2-limited and above Pcorner blocker– IP3-limited (Fig. 2.16).

(37)

iIP3= 0 dBm and iIP2= +20 dBm Pcorner blocker of −40 dBm is obtained. This is still below

all of the blockers considered here, so according to the calculation the front-end is clearly IP3-limited. Fig. 2.17 shows the most of the spectral components coincide regardless of system

adequateIP2in the mixer.

It is useful to determine bounds on the necessary values of IP2 and IP3 in cognitive radios. A plausible approach

is to assume the intermodulation components resulting from second- and third-order nonlinearity have equal magnitudes for a certain input level in a two-tone test [Fig. 5(a)]. Denoting

f f₁ f₂ f₂− f₁ f₂+f₁ P_int P_IM f2 −f 1 2 f −f 2 12 P ∆ 1 2 3 P_intIP IP Fundamental IM2 IM3 P_in 3 2 P_IM (a) (b)

Fig. 5. (a) Input power level producing equal IM2 and IM3 products, (b)

illustration of IM2-limited and IM3-limited regions.

this level byPintand expressing the quantities in dB and dBm,

we have∆P =Pint,PIMand

∆P+Pint = IP2 (13) ∆P 2 +Pint = IP3: (14) Thus, 2Pint,PIM = IP2 (15) 3Pint,PIM = 2IP3: (16) That is, Pint=2IP3,IP2: (17)

For example, ifIP3 =,5 dBm andIP2 =+30 dBm, then Pint = ,40 dBm; i.e., the system tends to beIM2-limited

for interferers below this level andIM3-limited for interferers

above this level [Fig. 5(b)].

As with SDRs, the downconversion and upconversion mix-ing in cognitive radios must deal with the LO harmonics. As shown in Fig. 6(a) for the receive path, the harmonics of the LO can mix with interferers, corrupting the downconverted desired signal. Unlike SDRs, however, the decades-wide bandwidth of cognitive radios makes high-order LO harmonics still criti-cal. For example, an SDR operating in the range of 900 MHz to 5 GHz need deal with harmonics up to the fifth or sixth order whereas a CR accommodating the range of 100 MHz to 10 GHz must handle harmonics up to the 100-th order!

Recent work on SDRs has focused on harmonic-rejection mixers [9, 10] derived from the original concept in [11]. Il-lustrated in Fig. 6(b), the idea is to mix the RF signal with

f Desired Channel f f_LO 2f_LO 3f_LO Received Spectrum Spectrum LO (t ( 1 g (t ( g2 (t ( g₃ (t ( x ₂ y(t ( t (t ( 1 g ( t g (t ( g3 ( 2 −1 +1 (a) (b)

Fig. 6. (a) Effect of LO harmonics in a broadband receiver, (b) harmonic-rejection mixing.

multiple phases of the LO,g1(t)-g3(t), and sum the results

with proper weighting so as to cancel the effect of the third and fifth harmonics. It can be shown that ifx(t)g2(t)is scaled

by a factor of

p

2 with respect tox(t)g1(t)andx(t)g3(t), then

these harmonics are removed [11]. With typical mismatches, the effect of the harmonics is reduced by 30 to 40 dB.

If applied to cognitive radios, harmonic-rejection mixing faces several critical issues. First, even for third and fifth harmonics, it requires the generation and distribution of eight LO phases, a difficult task as the LO frequency reaches a few gigahertz (the maximum LO frequency whose harmonics prove troublesome). Second, harmonic mixing becomes very complex if harmonics of seventh and higher orders must be rejected. Third, this technique does not remove even LO har-monics that result from random asymmetries in the mixers or LO waveforms. Consider, for example, the single-balanced mixer shown in Fig. 7(a), withVOSmodeling theVGS

mis-match between M2 andM3. As illustrated in Fig. 7(b), the

resulting vertical shift in the LO waveform equivalently dis-torts the duty cycle of the switching ofM2andM3. It can be

shown that the second LO harmonic arising from this effect has a peak amplitude of

V2LO VLO 4 VOS VLO ; (18) 4

Figure 2.16: IP2 vs. IP3: the limiting Non-linearity Mechanism as Function of the Blocker Power [82] 102 ₁₀3 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −160 −120 −80 −40 0 IF [Hz] P ow er [dBm] Blocker RX Signal iIP2 = +60 dBm iIP2 = +40 dBm iIP2 = +20 dBm

102 ₁₀3 ₁₀4 ₁₀5 ₁₀6 ₁₀7 −160 −120 −80 −40 0 IF [Hz] P ow er [dBm] Blockers RX Signal iIP2 = +60 dBm iIP2 = +40 dBm iIP2 = +20 dBm (b) 1 Target, 2 Blockers

Figure 2.17: Various Targets/Blockers Scenario: IP2 swept, iIP3 = 0 dBm

IP2 rating (the performance differs at even mixing indices, where the intermodulation products are lower than those at the odd indices). Therefore the simulations also support equation 2.32: under given conditions IP3 (and not IP2) is the limiting parameter for FMCW operation with blockers.

(38)

2.2.4 Conclusions

Sharing of the same antenna between radar RX and TX (the so-called mono-static arrange-ment) is challenging due to the lack of RX-TX time domain multiplexing (inherent to FMCW radar principles). Simulations for two blockers _{−5 dBm each show that a hypothetical} front-end (which is assumed to have linearity of 20 dB higher than state-of-the-art front front-ends) exhibit artifacts masking the target by 60 dBc. Hence in the mono-static case front-end linearity im-provements are offset by the effects of the blockers. However, the caused spectral regrowth should allow a basic FMCW ranging functionality. The spectral regrowth range-domain indi-cations are to be spaced up to 100 mm apart (worst case of the furthest reflection, table 2.2). Even in case of harsh non-linearity, spectral regrowth amplitude coefficients gradually decrease in value [66], so one can assume a spectral regrowth band limited to_{≈ 10× the spacing between} the spectral components. This means that a range indication uncertainty would be around 1 m, still acceptable in automotive LRR application.

Additionally, the mono-static arrangement can be facilitated with one/several realizations of leakage-/reflection- power reduction circuitry ([68], [52], [62], [6]). A combination of those solutions addresses the drawbacks of the mono-static approach. Thus the aforementioned very strict front-end requirements are shifted onto leakage-/reflection- reducing circuitry and/or signal processing complexity.

The 77 GHz LRR front-ends, benchmarked in this work (table 5.4) could be also compared from applicability to a mono-static case where no leakage cancellers are deployed. While the low-linearity front-ends (such as [46], [106], [3], [73]) are compressed used in mono-static ar-rangement, the high-linearity front-ends (such as [97], this work) allow reasonable performance for targets distanced up to 200 meters, which is in-line with application’s requirements.

(39)

3 Circuit Environment

3.1 MMIC Process and Transistors

3.1.1 The Process of Choice: SG25H1 by IHP

Unlike other high performance processes, SG25H is a high performance BiCMOS technology family that is simplified and lacks the technologically complex epitaxially-buried, highly-doped sub-collectors and deep trench isolation [42]. The bipolar modules of SG25H are integrated into a 0.25 µm RF CMOS platform that needs 19 lithographic mask steps (SGB25VD). The CMOS platform offers 2.5 V MOS transistors for digital applications, an isolated NMOS device, an accumulation-type MOS varactor, a junction varactor. SG25H1 being the highest performance version of SG25H (the other two are complementary BiCMOS and high breakdown voltage versions, respectively). The main features of SG25H1 bipolar module are (the cross-section is shown in Fig. 3.1):

Five masks are needed for the HBT module.

The standard device has reduced dimensions and also lower parasitic capacitances than the H3 version – a minimum drawn area of 0.18× 0.84 µm2 _{results in corresponding f}

t,

fmax values of 180 GHz, 220 GHz.

An ft improvement compared to H3 version results from novel collector design, which

substantially reduces base-collector charging and transit times. The key device feature is the formation of the whole HBT structure in one active area, which allows collector resistance similar to the emitter resistance.

Shallow trench isolation only – the absence of deep trench isolation improves the heat dissipation and reduces the thermal resistance.

3.1.2 HBT Modeling

The Process Design Kit (PDK) provided VBIC models of the active devices (the equivalent circuit deployed in the VBIC model is shown in Fig. 3.2). VBIC is a bipolar junction transistor (BJT) model that was developed as a public domain replacement for the SPICE Gummel-Poon (SGP) model. VBIC is designed to be as similar as possible to the SGP model, yet overcomes its major deficiencies. VBIC improvements over SGP are [7]:

Improved Early effect modeling

(40)

Circuit Environment

„High-Performance“ HBT: X-section

Features of

SG25H1 bipolar module Only shallow trench isolation (STI)

HBT integration after gate module

Uniform active area Combination of selective and differential Si/SiGe:C/Si epitaxy

Drawn emitter area: 0.21x0.84 µm² (npn200) 0.18x0.84 µm² (npn201)

STI CoSi emitter contact

CoSi base contact

CoSi collector contact

p-_{Si substrate}

30nm boron doped Si_0.8Ge_0.2C_0.002

poly crystalline SiGe:C extrinsic base SiO₂ n doped poly crystalline Si emitter SiO₂ Trench SiO2 SIC n+ S/D n+ n doped Si collector 2 µm bipolar window

SIC: Selectively implanted collector S/D: Source/drain implant

Figure 3.1: IHP SG25H1 High-Performance HBT Cross-Section [29] Quasi-saturation modeling

Parasitic substrate transistor modeling Parasitic fixed (oxide) capacitance modeling Includes an avalanche multiplication model Improved temperature modeling

Base current is decoupled from collector current Electro-thermal modeling

Smooth, continuous model

3.2 Passive Elements

3.2.1 Stack

Nominal back-end option offers 3 thin metal (Al) layers, a MIM layer and a 2 µm thick metal (Al) layer. An additional optional 2nd_{thick 3 µm Al metal layer is deployed – Fig. 3.3) Together}

Advanced silicon MMICs for mm wave automotive radar front ends