Radioelectronics RF CMOS Transceiver Design

Transcription

1 Radioelectronics RF CMOS Transceiver Design courses/tsek26/ Jerzy Dąbrowski Division of Electronic Devices Department of Electrical Engineering (ISY) Linköping University 1 Objectives of the course Learn principles of wireless digital communication transceivers Gain knowledge of RF front-end circuits Learn design methods and techniques for RF circuit design in CMOS technology Understand the related possibilities and limitations 2

2 Organization of the course Lectures 12 x 2h Tutorials 6 x 2h (by Rashad Ramzan rashad@isy.liu.se ) Homework: optional Laboratory work 3 x 4h (by Rashad Ramzan and Henrik Fredriksson henfr@isy.liu.se) Course book: T.H.Lee, The Design of CMOS Radio- Frequency Integrated Circuits, Cambridge Univ.Press, 2004 supporting book: Bosco Leung, VLSI for Wireless Communication, Prentice-Hall, Pearson Education, 2002/2004 Examination: written exam 3 Outline of the lecture Wireless communication systems today RF transceiver at glance CMOS RF design issues - architectures - circuits and devices Summary 4

3 Bit Rate kb/sec 100,000 UWB Wireless Communication Systems Today 10, In-door WLAN Bluetooth Cordless Zigbee DECT PHS CT1/CT2 UMTS CDMA2000 GSM IS-54/IS-95 PDC 4G directions Paging 3G directions Cellular Also many other wireless applications and gadgets GPS Satellite 10m 100m 1000m 10km 100km 1000km Range 5 Overview of Standards Standard Access Scheme Frequency band (MHz) Channel Spacing Frequency Accuracy Modulation Technique Rate (kb/s) Peak Power GSM TDMA/ FDMA/ TDD (Tx) (Rx) 200 khz 90 Hz GMSK , 2, 5, 8 W DCS-1800 TDMA/ FDMA/ TDD (Tx) (Rx) 200 khz 90 Hz GMSK , 2, 5, 8 W DECT TDMA/ FDMA/ TDD khz 50 Hz GMSK mw IS-54 TDMA/ FDMA (Tx) (Rx) 30 khz 200 Hz π/4 QPSK , 1, 2, 3 W IS-95 CDMA/ FDMA (Tx) (Rx) 1250 khz N/A OQPSK 1228 N/A Bluetooth CDMA/ FDMA/FH khz 20 ppm GFSK ,4,100 mw b (DSSS) CDMA khz 25 ppm QPSK 1, 2, 11 Mb/s 1 W WCDMA (UMTS) W-CDMA/ TD- CDMA (Tx) (Rx) 5000 khz 0.1 ppm QPSK 3840 (max) 0.125, 0.25, 0.5, 2W 6

4 RF Transceiver at Glance RF Section Baseband Section RF Section analog, high frequencies Baseband Section - mostly digital today (DSP), low frequencies 7 Digital Transmitter Baseband signal ADC Modulation & DSP DAC Upconverter /Modulator Carrier Power Amplifier Digital baseband section (compression, coding, modulation, shaping ) RF section (up-conversion, filtering, power gain) Tradeoff between power efficiency and spectral efficiency 8

5 Digital Receiver RF Filter Down Converter ADC Demodulator & DSP DAC Low Noise Amplifier Carrier Baseband signal RF section (image rejection, low noise gain, down conversion, channel selection) Digital baseband section (equalization, demodulation, decoding, decompression) 9 CMOS RF design issues Disciplines in RF design Key goals for ICs for RF transceiver implementation Why CMOS technology? Old and new receiver architectures for CMOS Circuits and devices 10

6 Disciplines required in RF system design RF Design 11 RF Circuit Design Octagon Multi-objective approach RF Design In digital design only one main trade-off between speed and power 12

7 Ultimate objective Single-chip transceiver Minimum external components Inductors and capacitors integrated on chip Duplexer or switch RF Section Baseband DSP & Ctrl Baseband input/output Crystal Battery or power supply 13 Bluetooth CMOS TRx from Alcatel (2001) Low-IF Rx and quadrature Tx RF front-end Layout of single chip TRx (first commercial with integrated BB and ARM processors + memory) 14

8 WLAN CMOS TRx from Intel Intel RFIC transceiver on 0.18 µm TSMC CMOS technology (Taiwan Semiconductor Manufacturing Corporation). This IEEE a (in 5 GHz band) transceiver employs a directconversion architecture and includes an internal synthesizer. This is Intel's first RFIC used in a WLAN product. 15 Why CMOS Technology Submicron MOSFETs, 180,130, 90 nm today, very fast, f max >100GHz, perform well up to 10 GHz or more Good linearity for higher signal swing With multiple metal layers good capacitors and inductors (Q L up to 20) can be integrated on a chip Upper metal layers far from Si substrate reduce substrate losses Lower substrate doping helps to isolate RF blocks and reduce losses Large digital bocks (DSP & control) can be integrated on one chip (but introduce substrate noise), Bipolar process not suitable CMOS on Si cheaper from other technologies (BiCMOS, or CMOS on GaAs, SiGe) Many successful RF CMOS designs performed recently 16

9 Old & New receiver architectures Superheterodyne receiver /double conversion (good sensitivity and selectivity, good image rejection but tradeoffs) RF Filter LNA IR Filter IF Filter IFA LP Filter ADC LO1 I LO2 Q BB signal Discrete IR and IF filters not amenable for integration Low impedance of those filters raise power dissipation in LNA and first mixer (matching for off-chip needed) 17 Superheterodyne receiver (cont d) Receiver band B w f IF RF filter selects band, rejects off-band signals, f LO,k-1 f LO,k f k-1 f k f IR filter rejects off-band products, it has same band as RF filter LO1 frequency is adjusted to select the channel for downconversion IF filter selects channel, adjacent channels are partly suppress f IF f Constant intermediate frequency f IF 18

10 Old & New receiver architectures (cont d) Direct receiver (homodyne) (fewer components, image filtering avoided no IR and IF filters) RF Filter LNA DC removal + LPF ADC I Q LO Leakage f LO = f RF f IF = 0 LO Large DC offset can corrupt weak signal or saturate LNA (LO mixes itself), notch filters or adaptive DC offset cancellation eg. by DSP baseband control Flicker noise (1/f) be difficult to distinguish from signal Channel selection with LPF, easy to integrate, (noise-linearity-power tradeoff are critical, even-order distortions low-freq. beat) 19 Direct receiver (cont d) Receiver band B w f LO,k = f k RF filter selects band, rejects off-band signals, f k-1 f k f f IF = 0 f Wanted channel is corrupted by its mirror, IQ downconversion is needed to separate them with Hilbert transform LP filter selects channel, It is also anti-alias filter for ADC f IF = 0 f Useful for wideband systems, DC and 1/f noise can be removed by HPF 20

11 Old & New receiver architectures (cont d) Low-IF receiver RF Filter LNA DC notch + LPF ADC I Q LO Leakage LO Polyphase filter supports IQ rejection DC offset and 1/f do not corrupt the signal, like in the superheterodyne, still DC offset must be removed /saturation threat But image problem reintroduced / close image! Still even-order distortions can result in low-freq. beat differential circuits useful 21 Low-IF receiver (cont d) RF band filter In-band image channel f IF f LO,k f IF f k Desired channel f Close-image problem Image and desired channel signal overlap at f IF frequency, but due to I and Q paths and Hilbert transform the image can be suppressed Tough requirements for IQ match if image is large, otherwise signal strongly corrupted f IF = ½ BW ch typical More severe problem than in zero-if since the image can be much stronger than the signal. Good for GSM std. since the adjacent channel only 9dB larger, so rejection of db enough 22

12 Need for IQ down conversion Inherent mirror spectrum Down conversion to zero with one mixer Down conversion to zero with quadrature IQ mixer -ω 0 0 ω 0 -ω 0 0 ω 0 -ω 0 0 ω 0 Aliasing by mirror IQ mirror cancellation after using Hilbert transform and 0 0 π j sgn( ω ) 2 Y (ω) e ± Y QLP ILP (ω) 23 Need for IQ down conversion (cont d) Example: s FSK (t) = cos(ω IF t + ϕ(t)) dϕ(t)/dt = k x BB (t) s 1 (t) = cos(ω IF ( t-t 1 ) + ϕ(t-t 1 )) = cos(ω IF t ϕ(t-t 1 )) = sin(ω IF t + ϕ(t-t 1 )) y(t) = sin(ϕ(t) - ϕ(t-t 1 )) + sin(2ω IF t + ϕ(t) + ϕ(t-t 1 )) y out (t) = sin(ϕ(t) - ϕ(t-t 1 )) t 1 dϕ(t)/dt = kt 1 x BB (t) s FSK Quadrature detection 90 0 But if ω IF = 0: s 1 y LPF y out (t) = sin(ϕ(t) - ϕ(t-t 1 )) + sin(ϕ(t) + ϕ(t-t 1 )) y out so the detection cannot be achieved since the other component is folded 24

13 Direct conversion transmitter Base band Asinω c t I Q LO Acosω c t Up-conversion is performed in one step, f LO = f c Leakage of LO BPF Simple modulation, e.g. QPSK can be done in the same process BPF suppresses harmonics LO must be shielded to reduce corruption I and Q paths must be symmetrical, otherwise crosstalk Leakage of PA PA Matching Network Receiver Also effect on Rx can be critical High-power signal Duplexer or Switch FDD or TDD, respectively 25 Problem of carrier leakage Q Asinω c t Leakage of LO Wanted transmitted signal Carrier falls in band I Acosω c t to BB BPF PA Calibration feedback f RF Constellation can be destroyed and EVM rises Tx measures output when BB signal is absent and introduces offset in BB stage to compensate for the carrier leakage 26

14 Circuits and devices A suspended spiral inductor of about 100 nh on a heavily doped CMOS substrate. This element enabled the first RF CMOS circuits at 900 MHz. Without inductors you can hardly neutralize parasitic capacitances in CMOS Today lightly doped substrates and uppermetal layers enable good inductors on chip Low-noise amplifier styles commonly used in CMOS. (a) Common gate circuit, robust against parasitics, moderate noise figure. (b) Common source circuit, lowest noise figure. 27 Circuits and devices (cont d) Differential style must be preferred to suppress common mode substrate noise and even order distortions CMOS oscillators require good Q factors to suppress own noise (white and 1/f) L Shield helps to reduce substrate currents But still eddycurrents 28

15 Circuits and devices (cont d) Circuit techniques of minimizing Phase Noise, 2nd harmonic can be suppressed and less 1/f noise is upconverted in this way 29 Circuits and devices (cont d) CMOS process tolerances can be large but a circuit correction is possible DfC MOSFETs as switches enable discrete tuning to enhance freq. range and compensate for production variations (tolerances). The same technique is popular to calibrate for IQ mismatch 30

16 Summary Wireless communication systems (mobile, cordless, WLAN, GPS, ) are in continuous progress Wireless communication systems are very complex multidisciplinary field Design of RF IC s is a multi-objective task CMOS technology proves to be increasingly competitive for RF IC s design (even higher frequencies) RF CMOS is an attractive research field 31