IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO.

5, MAY 1999

573

A 900-MHz/1.8-GHz CMOS Transmitter for Dual-Band Applications

AND

I. INTRODUCTION HE availability of new frequency bands in the 1.8-GHz range and the proliferation of various wireless standards have motivated vigorous efforts in the area of multistandard transceivers. Providing both higher exibility and roaming capability and increasing the overall capacity in mobile communications, dual-band operation has rapidly penetrated various radio-frequency products. However, cost and form factor considerations severely constrain the choice of the architecture and frequency planning as well as the design of the building blocks of such transceivers. In particular, the number of oscillators, frequency synthesizers, and external lters and resonators must be minimized. This paper describes the design of a 900-MHz/1.8-GHz CMOS transmitter for dual-band applications, with emphasis on compatibility with the Global System for Mobile Communication (GSM) and Digital Communication System at 1800 MHz (DCS1800). Employing two upconversion steps, the circuit generates 900-MHz and 1.8-GHz outputs that can assume linear or nonlinear modulation depending on the type of signals applied to the baseband ports. A prototype fabricated in a digital 0.6- m CMOS technology displays unwanted spurs 40 dB below the carrier while consuming 75 mW from a 3-V supply. To our knowledge, this is the rst published work on a dual-band transmitter. Section II of the paper deals with general issues in dualband transmitters. Section III presents the transmitter architecture, and Section IV describes the design of the building blocks. Section V summarizes the experimental results.
Manuscript received September 7, 1998; revised December 18, 1998. The author is with the Department of Electrical Engineering, University of California, Los Angeles, CA 90095 USA (e-mail: [email protected]). Publisher Item Identier S 0018-9200(99)03673-2.

Fig. 1. Time offset between receive and transmit time slots.

II. GENERAL CONSIDERATIONS In this design, the GSM and DCS1800 standards have been chosen as the framework. These standards incorporate the same modulation format, channel spacing, and antenna duplexing. Table I summarizes the characteristics of each standard, indicating that a dual-band transceiver can exploit the properties common to both so as to reduce the off-chip hardware. To minimize the number of oscillators and synthesizers, the receivers and transmitters in a dual-band system must be designed concurrently, with the frequency planning chosen such that the receive and transmit paths are driven by the same synthesizers. Although GSM and DCS1800 use frequency-division duplexing (FDD) at the front end, their actual operation is somewhat similar to time-division duplexing (TDD) because their receive and transmit time slots are offset by 1.73 ms (three time slots) (Fig. 1). Thus, frequency synthesizers can be time-shared between the receiver and the transmitter. The dual-band transmitter described herein is designed in conjunction with the dual-band receiver reported in [1]. To address the frequency planning issues, we briey look at the receiver (Fig. 2). Based on the Weaver image-reject architecture [2], the receiver performs the rst downconversion such that the GSM and DCS1800 bands appear as images of each other. The Weaver topology then selects one band and rejects the other by addition or subtraction of the spectra and . The rst local oscillator (LO) frequency at points

00189200/99$10.00 1999 IEEE

574

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 5, MAY 1999

Fig. 4. Simple dual-band transmitter.

Gaussian lter, thereby generating smooth edges at the input of the frequency modulator. The resulting output is expressed as
Fig. 2. Dual-band receiver architecture.
BB

(2)

(a)

is the impulse response of the where Gaussian lter. The conceptual method described by Fig. 3(b) and (2) is indeed employed in some transmitters, e.g., for the Digital European Cordless Telephone standard. However, if the amplitude of the baseband signal applied to the VCO or the gain of the VCO are poorly controlled, so is the bandwidth of the modulated signal. For this reason, in high-precision systems such as GSM, the waveform in (2) is rewritten as (3) , and and where are generated by accurate mixed-signal techniques [5], [6]. Equation (3) forms the basis for our transmitter design. III. TRANSMITTER ARCHITECTURE To employ 450- and 1350-MHz LO frequencies, we postulate that the transmitter must incorporate two upconversion steps: from baseband to an intermediate frequency (IF) of 450 MHz and from 450 to 900 MHz or 1.8 GHz. We also recognize that a simple mixer driven by the 450-MHz IF and the 1350-MHz LO generates the 900-MHz and 1.8-GHz signals with equal amplitudes, necessitating substantial ltering to suppress the unwanted component. It is therefore desirable to perform the second upconversion by single-sideband (SSB) mixing. With the foregoing observations, we consider the topology shown in Fig. 4 as a possible solution. The baseband and signals are upconverted to 450 MHz and subsequently separated into quadrature phases by means of an network, resulting in and . Single-sideband mixing of the IF and the second LO signals is then carried out by two mixers, with their outputs added or subtracted so as to produce the 900or 1800-MHz output according to the band select command. The architecture of Fig. 4 provides a compact solution for dual-band operation, but it suffers from several drawbacks. network introduces a loss of 3 dB in the First, the signal path, and, more important, loads the rst upconverter. Second, both of the outputs appear at the same port, making it difcult to utilize narrow-band tuned amplication at this

(b) Fig. 3. Generation of FSK and GMSK signals.

is therefore equal to 1350 MHzmidway between the two bandsand the second LO frequency is in the vicinity of 450 MHz. To minimize the number of LOs and synthesizer loops, it is desirable to utilize the same frequencies for the transmit path as well. Before considering suitable transmitter architectures, we review Gaussian minimum shift keying (GMSK) modulation to arrive at some of the design implications. Fig. 3 conceptually illustrates frequency shift keying (FSK) and GMSK. In FSK, rectangular baseband pulses are directly applied to a frequency modulator, e.g., a voltage-controlled oscillator (VCO), thereby creating an output waveform given by
BB

(1)

is a constant denoting the depth of modulation where represents the baseband signal. and An important drawback of FSK is the large bandwidth occupied by the modulated signal, partly because of the abrupt transitions in the frequency introduced by the sharp edges of the baseband pulses. We expect that if the frequency changes more smoothly from one bit to the next, then the required bandwidth decreases. In fact, the spectrum of the signal , where expressed by (1) decays in proportion to denotes the highest continuous derivative of [3]. Based on this observation, GMSK modulation alters the shape of the baseband pulses so as to vary the frequency gradually. As shown in Fig. 3(b), the rectangular pulses are rst applied to a

RAZAVI: 900-MHz/1.8-GHz CMOS TRANSMITTER

575

Fig. 5. Effect of unwanted sideband on DCS1800 output.

Fig. 7. Upconversion with quadrature outputs.

Fig. 6. Increase in adjacent-channel power due to second harmonic of unwanted sideband.

port. Third, even with perfect matching between the quadrature paths in the SSB mixer, the variation of the absolute value with process and temperature leads to considerable of and , thereby creating a amplitude mismatch between signicant unwanted sideband at the output. For example, a results in an unwanted sideband only 20 dB 20% error in below the wanted component. The existence of an unwanted sideband 900 MHz away from the desired signal may seem unimportant because various ltering operations in the following power amplier (PA) and matching network provide further suppression. However, second-order distortion in the PAa signicant effect because PAs are typically single endedmay lead to a troublesome phenomenon in the generation of DCS1800 signals. Illustrated in Fig. 5, the issue arises because the second harmonic of the 900-MHz sideband falls in the transmitted DCS1800 channel. Since, from Carsons rule [4], the second harmonic of a frequency-modulated signal occupies roughly twice as much bandwidth as the rst harmonic, the 1800-MHz output may exhibit substantial adjacent-channel power, violating the transmission mask (Fig. 6). Thus, the unwanted sideband produced by the SSB mixer must be sufciently small. In summary, the architecture of Fig. 4 requires two modications: 1) the IF quadrature generation must avoid the use of networks and 2) the GSM and DCS1800 paths must be separated at some point such that each can incorporate narrow-band tuning. To produce the quadrature phases of the 450-MHz IF signal, we recognize that the baseband signal is available in and in (3). The IF quadrature phases, namely, signal can thus be generated in quadrature form as depicted in Fig. 7, where proper choice of the phases together with addition or subtraction at the output yields both and . Compared to the circuit of Fig. 4, this conguration both provides higher gain balance between the two

Fig. 8. Two-step upconversion generating 900-MHz and 1.8-GHz outputs.

paths and avoids the loss and loading of the network while using two more mixers. The additional mixers consume more power, but since the conversion gain is higher in this case, the following SSB mixers require less power, leading to an overall power dissipation comparable to that of Fig. 4. The 450-MHz outputs of the rst upconverter can now be multiplied by the quadrature phases of the second LO and added or subtracted to generate the GSM and DCS1800 signals. As mentioned above, it is preferable to design the second upconversion modulators such that the 900-MHz and 1.8-GHz waveforms appear at the outputs of two different circuits, thus allowing efcient narrow-band amplication. This is accomplished as shown in Fig. 8, where two independent SSB modulators produce the two bands according to the band select command. Note that narrow-band tuning also suppresses the unwanted sideband resulting from mismatches in the SSB mixers. To save power consumption, only one of the modulators is active in either mode. The overall architecture of the dual-band transmitter is and shown in Fig. 9. Since all of the signals up to ports are differential, each band incorporates a differential to

576

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 5, MAY 1999

Fig. 9. Dual-band transmitter architecture. (a)

single-ended (D/SE) converter, applying the result to an output buffer. The transmitter requires external power ampliers to deliver the high power levels specied by GSM and DCS1800. IV. BUILDING BLOCKS In this section, we describe the transistor-level implementation of each building block, emphasizing the design constraints imposed by the architecture. The circuit topologies are identical for both bands, but device dimensions and bias currents are chosen to optimize the performance of each. A. First Upconversion The 450-MHz upconversion modulator consists of two Gilbert cell mixers whose outputs are added in the current domain. Shown in Fig. 10, the circuit utilizes resistive source degeneration, thereby improving the linearity in the baseband port of each mixer. Two 100-nH inductors convert the output current to voltage. To minimize the area occupied by each inductor, a stack of three spiral structures made of three metal layers [Fig. 10(b)] is used [1], reducing the area by approximately a factor of eight [7]. Since the polysilicon connection and the bottom spiral suffer from substantial parasitic capacitance to the substrate, this node is connected to the supply voltage, increasing the self-resonance frequency of the inductor. The quality factor of the inductor is estimated to be about four, and the self-resonance frequency is about 600 MHz. Why must the baseband ports be linearized? Let us return to and experience third(3) with the assumption that order distortion. The resulting IF signal can then be expressed as (4) represents the amount of third-order nonlinearity. where Grouping the terms in (4), we obtain
(b) Fig. 10. (a) First upconversion modulator and (b) implementation of load inductors.

Fig. 11. Simulated spectra of the two terms in (5) (horizontal scale normalized to bit rate; vertical scale 5 dB/div.).

(5) Equation (5) reveals that third-order distortion gives rise to a

component centered around but with a modulation index three times that of the ideal GMSK signal. Invoking Carsons rule, we postulate that the second term occupies roughly three times the bandwidth, raising the power transmitted in adjacent channels. Fig. 11 shows the simulated spectra of , indicating that the the two components in (5) with

RAZAVI: 900-MHz/1.8-GHz CMOS TRANSMITTER

577

Fig. 12.

Effect of harmonic distortion at baseband ports. Fig. 16. Output buffer.

Fig. 13.

Voltage-to-current converter with output switching.

Fig. 14.

Single-sideband modulator circuit.

Fig. 15. Differential to single-ended conversion using (a) tuned current mirror and (b) negative resistance generator.

Fig. 17. Transmitter die photograph.

unwanted signal indeed consumes a wider band. For this reason, as depicted in Fig. 12, must be small enough that the transmission mask is not violated. In this design, the resistive degeneration and tail currents are chosen so as to with a 0.5-V baseband input, yielding a ensure third-order component 40 dB below the desired signal. B. SSB Modulator and in Fig. 10 must The signals generated at nodes be routed to one of the SSB modulators according to the

band select command. As illustrated in Fig. 13, the routing is performed in the current domain to minimize signal loss due to addition of the switches. Capacitively coupled to the output of the 450-MHz upconverter, the voltage-to-current converter employs grounded-source input devices to save the voltage headroom otherwise consumed by a tail current source. The and . Note bias current of the circuit is dened by operate in the deep triode region, sustaining a that small voltage drop. Also, the linearity of this and subsequent stages is not critical because GMSK signals display a constant envelope and are quite insensitive to spectral regrowth [6].

578

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 34, NO. 5, MAY 1999

(a)

(b) Fig. 18. Measured output spectrum at (a) 898 MHz and (b) 1.8 GHz (horizontal 1 MHz/div.; vertical 10 dB/div.).

With the quadrature phases of the IF signal available in the current domain, SSB mixing assumes a simple topology. Shown in Fig. 14, the circuit senses the differential current signals routed from each 450-MHz upconverter, performs mixing with the second LO, adds the resulting currents with proper polarity, and converts the output to single-ended form. C. Differential to Single-Ended Converter To achieve a reasonable gain, it is desirable to employ tuning in the D/SE converter. Fig. 15 depicts two realizations of such a circuit. In Fig. 15(a), a current mirror together with two

and , reducing the inductors creates resonance at nodes effect of device capacitances. The difculty here is the large gate-source capacitance of the two PMOS devices, mandating and hence a low conversion gain. a small value for Fig. 15(b) presents an alternative topology where a PMOS device introduces a negative resistance in parallel with a oating inductor. It can be shown that (6) where and denote the total capacitance at nodes

RAZAVI: 900-MHz/1.8-GHz CMOS TRANSMITTER

579

and , respectively [6]. As a compromise between margin to oscillation and boost in gain, the negative resistance, , is chosen to increase the of the inductor can assume by approximately a factor of two. Note that and in series. a relatively large value because it sees because this port In this design, the signal is sensed at exhibits a lower output impedance. Simulations indicate that the topology of Fig. 15(b) provides about three times the voltage gain of the circuit in Fig. 15(a). D. Output Buffer The output buffer is shown in Fig. 16. Two common-source and , boost the signal level, driving the 50stages, impedance of the external instrumentation. The bias current is dened by and , and that of by and of . Neglecting the dc drop across the inductor, we have ; that is, can be ratioed with respect to . V. EXPERIMENTAL RESULTS The dual-band transmitter has been fabricated in a 0.6- m CMOS technology. Fig. 17 shows a photograph of the die, 850 m . All of the inductors are which measures 1300 integrated with no additional processing steps. The prototype is directly mounted on a printed circuit board and tested with a 3-V supply. The baseband signal consists of quadrature phases of a 1.2-MHz sinusoid, and the quadrature phases of LO and LO are generated by external couplers. Fig. 18(a) shows the measured output spectrum when the 900-MHz band is selected. In addition to the desired component, the spectrum exhibits spurs corresponding to , the lower sideband (resulting from mismatches in the rst upconverter), and upconverted harmonics of the baseband signal. Similarly, as shown in Fig. 18(b), the 1.8-GHz output , the upper sideband, and contains spurs at harmonics of the baseband input. Note that in both cases, the spurious components are at least 40 dB below the carrier. The circuit dissipates 75 mW, of which 10 mW is drained by the four upconversion mixers, 10 mW by the SSB mixer, and 55 mW by the output buffer. Two aspects of these results should be revisited for GSM/DCS1800 applications. First, the output level is relatively low, necessitating an additional buffer prior to the power amplier. Second, from the plots of Fig. 18(a) and (b), the thermal noise at large frequency offsets is approximately

PERFORMANCE

OF

TABLE II DUAL-BAND TRANSMITTER

equal to 107 dBc/Hz at 900 MHz and 104 dBc/Hz at 1.8 GHz, requiring that the front-end duplexer lter provide adequate suppression of this noise in the receive band. Table II summarizes the performance of the dual-band transmitter. VI. CONCLUSION The design of dual-band transmitters poses many challenges in terms of frequency planning and compatibility with their corresponding receivers. A two-step transmitter architecture has been introduced that provides dual-band operation with 450- and 1350-MHz LO frequencies. Also, circuit techniques for the generation of the quadrature phases of the IF signal as well as differential to single-ended conversion are presented. ACKNOWLEDGMENT The author wishes to thank R. Thiara for simulation results of Fig. 11. REFERENCES
[1] S. Wu and B. Razavi, A 900-MHz/1.8-GHz CMOS receiver for dualband applications, IEEE J. Solid-State Circuits, vol. 33, pp. 21782185, Dec. 1998. [2] D. K. Weaver, A third method of generation and detection of singlesideband signals, Proc. IRE, vol. 44, pp. 17031705, Dec. 1956. [3] J. B. Anderson, T. Aulin, and C.-E. Sundberg, Digital Phase Modulation. New York: Plenum, 1986. [4] L. W. Couch, Digital and Analog Communication Systems, 4th ed. New York: Macmillan, 1993. [5] K. Feher, Wireless Digital Communications. Englewood Cliffs, NJ: Prentice-Hall, 1995. [6] B. Razavi, RF Microelectronics. Upper Saddle River, NJ: PrenticeHall, 1998. [7] R. B. Merril et al., Optimization of high Q inductors for multi-level metal CMOS, in Proc. IEDM, Dec. 1995, pp. 38.7.138.7.4.

Behzad Razavi (S87M90), for a photograph and biography, see p. 276 of the March 1999 issue of this JOURNAL.