RFIC’s Challenges for Third Generation Wireless Systems

Olga Boric-Lubecke*a, Jenshan Lin**b, Penny Gouldc, and Munawar Kermallida Bell Labs, Lucent Technologies, 600 Mountain Ave., Murray Hill, NJ 07974

bAgere Systems, 600 Mountain Ave., Murray Hill, NJ 07974cBell Labs, Lucent Technologies, Swindon, UKdBell Labs, Lucent Technologies, Whippany, NJ

ABSTRACT

Third generation (3G) cellular wireless systems are envisioned to offer low cost, high-capacity mobile communicationswith data rates of up to 2 Mbit/s, with global roaming and advanced data services. Besides adding mobility to theinternet, 3G systems will provide location-based services, as well as personalized information and entertainment. Lowcost, high dynamic-range radios, both for base stations (BS) and for mobile stations (MS) are required to enableworldwide deployment of such networks. A receiver’s reference sensitivity, intermodulation characteristics, andblocking characteristics, set by a wireless standard, define performance requirements of individual components of areceiver front end. Since base station handles multiple signals from various distances simultaneously, its radiospecifications are significantly more demanding than those for mobile devices. While high level of integration hasalready been achieved for second generation hand-sets using low-cost silicon technologies, the cost and size reduction ofbase stations still remains a challenge and necessity. While silicon RFIC technology is steadily improving, it is stilldifficult to achieve noise figure (NF), linearity, and phase noise requirements with presently available devices. Thispaper will discuss base station specification for 2G (GSM) and 3G (UMTS) systems, as well as the feasibility ofimplementing base station radios in low-cost silicon processes.

Keywords: 3G, wireless, base stations, silicon RFIC

1. INTRODUCTION

Second generation (2G) digital wireless systems, such as Global System for Mobile Communications (GSM) andcdmaOne (IS-95), have been widely deployed over the last decade, resulting in over 800 million users worldwide, withabout additional 60 million of analog (first generation) users. Number of mobile phones exceeds the number of landlinephones and the mobile phone penetration exceeds 70% in countries with the most advanced wireless markets. Eventhough this growth has recently somewhat slowed down, demand for more data handling capability is still there. Thirdgeneration (3G) standards, envisioned to provide enhanced capacity, quality and data rates, are currently beingdeveloped across the industry and by global groups such as the Third Generation Partnership Project (3GPP). Theenhancements offered in 3G services will enable high speed multimedia and internet access. Universal MobileTelecommunication System (UMTS), to replace GSM, has emerged as the most widely adopted third generation airinterface. Its specifications has been created in the 3rd Generation Partnership Project (3GPP), which is the joint projectof the standard bodies from Europe, Japan, Korea, USA, and China. Deployment of the first UMTS network is wellunderway in Japan by NTT DoCoMo, and will also become available in Europe in 2002.

*olga@lucent.com; phone +1 908 582 1889; fax +1 908 582 4941; Bell Labs, Lucent Technologies, 600 Mountain Ave., Murray Hill,NJ 07974, USA; **jenshan@agere.com; phone +1 908 582 5182; fax +1 908 582 4941; Agere Systems, 600 Mountain Ave.,Murray Hill, NJ 07974, USA

Base transceiver station (BTS) radio requirements are more stringent that those for user interface units for all cellularwirelles standards. While handsets handle only one channel at a time, a base station receiver handles multiple channelsand thus requires a higher dynamic range [1]. Mobile receivers that meet GSM and other standards requirements havealready been implemented in low cost silicon technology [2], while base station receivers are typically still realized inGaAs technology [3]. Even though silicon RFIC technology is steadily improving, it is still a challenge to achieve noisefigure (NF), phase noise, and linearity requirements for BTS applications with presently available devices. However,with careful system specification partitioning and choice of radio architecture, it was demonstrated that it is possible tomeet GSM900 and DCS1800 BTS receiver requirements in BiCMOS technology [4,5]. While 3G systems promise asubstantially enhanced services, radio specifications will be similar to those of 2G systems, and it should be possible tomeet them in silicon technologies as well.

A receiver’s reference sensitivity, intermodulation characteristics, and blocking characteristics define performancerequirements of individual components of a radio front end. While direct downconversion architecture has been exploredfor handsets [6], base stations typically use heterodyne receiver front-end architecture, as shown in Figure 1. It would bedesirable to have as many of these radio components as possible integrated in silicon, to reduce both the cost and size ofthe base stations. While there is no compact, integrated solution for filtering components available today, othercomponents could in principle be integrated on a single chip.

In this paper, we will derive radio specifications for a 2G (GSM) and a 3G (UMTS) standard, to illustrate that basestation requirements are more stringent than for handsets, and to show that 2G and 3G requirements do not differ greatly.Next, we will discuss silicon implementaion of most important receiver circuits: low noise amplifier (LNA), mixer, andvoltage-controlled oscillator (VCO). Finally we will discuss the feasibility of implementing a GSM BTS receiver insilicon-MMIC 0.25 µm BiCMOS process. The receiver chips that achieve low noise figure and high third order intercept(IP3) simultaneously without any gain control will be described. These chips were fabricated using a 0.25µm BiCMOSprocess with inductor quality factor (Q) of approximately 10. With a supply voltage of 3V, current consumption is132mA for the GSM900 receiver and 117mA for the DCS1800 receiver. This was the first reported silicon-integratedradio front end for base stations [4].

Fig. 1 Typical BS receiver front-end architecture.

2. UMTS AND GSM STANDARDS

GSM is currently the most widely used digital cellular radio system. As of July 2001 [7], there were over 560 millionGSM users, accounting for two thirds of world users. In contrast, second most popular standard, cdmaOne, has onlyabout 100 million users. GSM provides international roaming capability, which is currently available in over 150countries. GSM standard is based on a time-division-miltiple access (TDMA) and frequency-division-multiple access

VCO

BPF SAWMixerLNA IF Amp

Duplexer

To AntennaSynthesizer

From Transmitter

To IF Subsystem

BufferAmp

(Interchangeable)

Driver

(FDMA) scheme, and it uses Gaussian Minimum Shift Keying (GMSK) modulation. The bit rate is 270 kbit/s, andchannel bandwidth 200 kHz, with each channel shared by 8 users in a TDMA mode. It uses a full duplexcommunication, in three frequency bands: 900 MHz, 1800 MHz, and 1900 MHz. Digital cellular system (DCS) 1800MHz band was introduced in Europe in early 1990’s to provide more bandwidth. Personal communications services(PCS) 1900 MHz band is used in the US. GSM and UMTS bandwidth allocations and other features are summarized inTable I.

Table I. GSM and UMTS features

GSM900 DCS1800 PCS1900 UMTS

MS Tx-BTS Rxf [MHz]

890-915 1710-1785 1850-1910 1920-1980

BTS Tx-MS Rxf [MHz]

935-960 1805-1880 1930-1990 2110-2170

Channelspacing [MHz]

0.2 0.2 0.2 5

Users per cell 992 2992 2383 3060

Modulation GMSK GMSK GMSK QPSK

MS TxPower [W]

2 1 1 0.25

GSM is a second generation cellular standard, which will gradually be replaced by a third generation UMTS standard.UMTS, or WCDMA (wide-band CDMA) standard is based on direct sequence spread spectrum technology, with codedivision multiple access (CDMA) scheme. User information bits are spread over a wide bandwidth by multiplying theuser data with quasi-random bits (called chips) derived from CDMA spreading codes. All users in one cell share samefrequencies all the time, and signal is “spread” over the whole bandwidth of a channel. This “spreading” of the signalresults in a processing gain, allowing for the detection of signal levels below the noise floor. UMTS uses QuaternaryPhase Shift Keying (QPSK) modulation, with the chip rate of 3.84 Mcps, and corresponding channel bandwidth of about5 MHz. Data rates are variable, depending on the type of service. For example, voice service will operate at 12.2 kbps,while non-real time data service will operate at 384 kbps. NTT DoCoMo’s Freedom of Mobile multimedia Access(FOMA) service, to begin in Tokyo area by the end of 2001, is using a downlink data rate of 384 kbps. One disadvantageof the CDMA type systems is that they require elaborate transmit power control for mobile units. Since all users in thecell use the same bandwidth simultaneously, if the transmit power is not controlled, a single strong user close to the basestation can completely mask all the users far from the base station. Receiver power from all users must be within 1dB,and mobile power must be able to vary by as much as 85 dB in a few microseconds.

3. GSM SPECIFICATIONS

GSM base station and mobile radio receiver requirements, referenced at the antenna connector, will be derived from theGSM specification 05.05 document [1]. Values obtained for noise figure (NF), linearity and phase noise should beconsidered as reference values only. Vendors typically design their products to exceed these values by a certain marginto give them a competitive advantage.

3.1 Noise FigureReceiver NF can be determined from reference receiver sensitivity (S) [1], knowing the input noise floor, and signal tonoise ratio (SNR) requirement in the absence of interferes. Noise floor (N) can be determined from the Nyquist equationfor thermal noise:

N = kTB, (1)

where k is a Boltzman constant, T is the temperature in Kelvins, and B is the single-sided noise bandwidth of the system.At room temperature, T=290K, noise power per bandwidth can be calculated to be –174 dBm/Hz. In the case of a GSMreceiver, channel bandwidth is 200 kHz, resulting in a noise floor of –121 dBm. The SNR depends on the bit-error-rate(BER) required for digital radio quality and the implementation of baseband algorithms, and can be calculated as [8]:

SNR= (Eb/No) x (R/B), (2)

where Eb is the energy per bit, No noise power density, R bit rate, and B channel bandwith. To achieve BER of 10-3,using GMSK modulation with frequency hoping, Eb/No of 7.6 dB is needed [8], resulting in SNR of 8.9 dB. For givenreference receiver sensitivity S, maximum noise figure can be calculated as:

NFmax = (S-SNR) –N. (3)

Table II shows the receiver sensitivity and NF requirement assuming SNR of 9 dB. Due to lower reference sensitivity, a2 dB lower NF radio is required for normal base stations than for mobile devices. Normal BTS has the lowest receiverNF specification of 8 dB at both bands.

Table II. Rx front-end NF requirements

GSM900Small MS

DCS1800MS

GSM900Micro BTS

DCS1800Micro BTS

GSM900Normal BTS

DCS1800Normal BTS

Ref. Sen[dBm] -102 -102 -97 -102 -104 -104NF [dB] 10 10 15 10 8 8

3.2 LinearityReceiver intermodulation characteristics determine linearity requirements (third-order intercept point or IP3) for thefront end. According to GSM standard [1], reference sensitivity performance should be met when the following threesignals exist simultaneously at the input of the receiver: (1) a wanted signal at frequency fo 3dB above the referencesensitivity, (2) a CW interfering signal at 800 kHz offset from fo, and (3) a modulated interfering signal at 1600 kHzoffset from fo, as shown in Fig. 2. Two interfering signals should be of the same strength (Table III).

Fig. 2 GSM intermodulation specification reffered to input (for normal BTS).

DCS1800: -49dBm

-101dBm

-113dBm

GSM900: -43dBm

Wanted

Intermodulationcomponent

Interferer 2Interferer 1

Pow

er

Frequencyf0 f0 +800 KHz f0 +1600 KHz

Fig. 3 Third-order intercept point calculation (GSM900 normal BTS). G is the receiver gain.

For example, GSM900 normal BTS reference sensitivity is –104 dBm. Therefore wanted signal should be at –101 dBm,and interferers are specified to be –43 dBm. Intermodulation component, combined with noise, should be SNR lowerthan the wanted signal. Assuming that intermodulation component and noise have the same power, we arrive at theconclusion that intermodulation component, referred to input, should be –113 dBm. Once interferer (Pt) andintermodulation (Pi) power levels are known, it is straight forward to calculate input IP3 (IIP3) , using diagram in Fig. 3[9]:

IIP3 = Pt + (Pt– Pi) /2. (4)

Interferer levels specified in [1], and corresponding IIP3 values are shown in Table III. Since there is less attenuation atthe air interface at 900 MHz than at 1800 MHz, 900 MHz BTS interferers are specified 6dB higher than for DCS 1800BTS, and resulting IIP3 requirement will be significantly more stringent.

Table III. Rx front-end input IP3 requirements

GSM900Small MS

DCS1800MS

GSM900Micro BTS

DCS1800Micro BTS

GSM900Normal BTS

DCS1800Normal BTS

Interferer[dBm]

-49 -49 -43 -49 -43 -49

IIP3[dBm]

-18 -18 -10 -18 -8 -17

3.3 Phase NoiseThe difficult phase noise (PN) requirement for the VCO comes from the in-band blocking characteristics. Wanted signalmay be accompanied by a large interferer in an adjacent channel, which will produce an unwanted component at thedownconverted frequency of the useful signal if the local oscillator exhibits finite phase noise (Fig. 3). When the twosignals are mixed with the LO output, the resulting signal consists of two overlapping spectra, with the wanted signal at

IP3

-43dBm-113dBm

slope

=1

slop

e=3

IIP3

OIP3

Pin

Pout

Pi+G

Pt

∆P

∆P/2

Pt+G

Pi

the same frequency as the “tail” of the interferer. To be able to detect the signal, unwanted component due to theblocker, Pi, must be at least SNR lower than the useful signal, P(IF) [9].

Fig. 4 Phase noise requirement derived from blocker specification for DCS1800 Micro BTS. CL stands for conversionloss of the mixer.

Since Pi can be calculated as:

Pi = P(LO-Blocker) +PN[dBc/Hz]+ 10logB, (5)

maximum phase noise can be found as:

PN [dBc/Hz] = P(RF)-SNR-P(Blocker)-10logB. (6)

Based on the blocking requirement, and the assumption that SNR is 9 dB, the phase noise specifications were calculatedand summarized in Table IV [7]. GSM900 micro and normal BTS have the most stringent phase noise requirement, as isthe case for IIP3 requirement. Phase noise specifications are over 10 dB lower for normal BTS than for MS at 1800MHz, and close to 20 dB lower at 900 MHz.

Table IV. VCO phase noise requirement [dBc/Hz]

Offset[kHz]

GSM900Small MS

DCS1800MS

GSM900Micro BTS

DCS1800Micro BTS

GSM900Normal BTS

DCS1800Normal BTS

600 -118 -116 -130 -121 -137 -128800 -118 -116 -140 -131 -147 -1381600 -128 -126 -140 -131 -147 -1383000 -133 -133 -140 -131 -150 -138

P(RF) = -99dBm

P(Blocker) = -30dBm

800 kHz

P(IF) = -99 dBm + CL

P(LO - Blocker) = -30dBm + CL

LO

800 kHz

|PN| dBc/Hz

Pi < P(IF) - SNR

4. UMTS SPECIFICATIONS

The requirements for a UMTS receiver are based on the technical specifications established by the 3rd GenerationPartnership Project (3GPP). In particular, document 3GPP TS 25.104 (UTRA FDD: Radio Transmission and Reception)[10] covers the RF requirements for the entire base station. As in previous section, values for noise figure, IIP3 andphase noise derived here should be considered as reference values only, keeping in mind that all vendors introducemargins on these values to differentiate performance of their products.

4.1 Noise FigureSection 7 of TS 25.104 document [10] defines the base station receiver characteristics. The receiver sensitivity, isspecified as –121 dBm for a 12.2 kbps channel. At this level, the BER is expected not to exceed 0.1%. To achieve thespecified BER, we assume that Eb/No of 5.2 dB is required, which includes any coding gain in the DSP [11]. To convertthe 5.2 Eb/No into an RF parameter of SNR, we use equation (2), where R is 12.2 kbps, and B is 3.84 MHz. In this case,R/B represents processing gain due to the signal spreading, which is 25 dB. SNR is calculated to be –20 dB, indicatingthat signal power can be 20 dB under the interference or thermal noise power, and the WCDMA receiver can still detectthe signal. This means that to achieve the desired BER of 0.1%, we require that the SNR be at least –20 dB. To calculatethe minimum signal level that can achieve this desired BER, we now have to calculate the noise floor. Using equation(1), noise floor for a 3.84 MHz channel is found to be –108 dBm. Therefore, since the noise floor is –108 dBm, and theminimum SNR –20 dB, the maximum sensitivity of the receiver is calculated to be –128 dBm. Since the referencesensitivity requirement for 3GPP is –121 dBm, the maximum noise figure that can be tolerated is 7 dB. This would raisethe noise floor from –108 dBm to –101 dBm. With the required –20 dB SNR, a signal level of –121 dBm could still bedemodulated with the required BER. For comparison, maximum noise figure for UMTS BTS receiver is 1 dB lower thanmaximum noise figure for normal GSM900 and DCS1800 BTS receivers, and therefore imposes a somewhat morestringent requirement on LNA noise figure.

4.2 LinearityThe linearity specification of the receiver is defined by intermodulation characteristic in TS 25.104, which states thatreference performance should be met when the following three signals are present at the BST antenna input: a –115 dBmwanted signal, a –48 dBm unmodulated interferer at 10 MHz offset, and a –48dBm modulated interferer with one code at20 MHz offset. Since SNR was calculated to be –20 dB, intermodulation component, combined with noise, can be up to20 dB higher than the wanted signal, for the receiver to be able to detect the signal. Assuming that intermodulationcomponent and noise have the same power, we arrive at the conclusion that intermodulation component, referred toinput, should be –98 dBm. Using equation (4), we can calculate IIP3 to be –23 dBm. This IIP3 requirement issignificantly more relaxed than that for GSM and DCS1800 BS receivers (Table III). In practice, it is typically desired tohave intermodulation component sufficiently low so that it does to raise the noise floor of the receiver. To achieve this,we can assume that intermodulation component reffered to input should be 10 dB lower than the noise floor, or –118dBm, resulting in IIP3 requirement of –13 dBm. However, this IIP3 is significantly more stringent than standardrequires.

4.3 Phase NoiseSince UMTS signal occupies the whole 5 MHz bandwidth, there are no in-band blockers which set-up stringent phasenoise requirements on local oscillator phase noise. Unlike GSM900 and DCS1800 BTS receivers, which have a verydemanding phase noise specification at 800 kHz and 1600 kHz offsets, UMTS receiver phase noise is determined by ablocker at a 10 MHz offset. According to the standard, receiver should be able to handle the wanted signal at the level of–115 dBm, in the presence of –40 dBm blocker 10 MHz away. Using equation (6), and assuming that phase noise isconstant over a 3.84 MHz bandwidth, we can calculate that phase noise should be better than –120.8 dBc/Hz at 10 MHzoffset (+/- 2.5 MHz). In addition, there is also spectrum emission mask specified for UMTS signals. This can besummarized as specifying that –14 dBm/30 kHz is required at 2.5 MHz offset and –13 dBm/1 MHz is required at 3.5MHz offset for BTS operating with a maximum output power of 43 dBm. Assuming this maximum output power, thesetwo specifications correspond to a phase noise requirement of –80 dBc/Hz at 2.5 MHz offset, and –110 dBc/Hz at 3.5MHz. This kind of phase noise is much less challenging that the GSM900 and DCS1800 specifications (Table IV), andcan easily be achieved with fully integrated silicon VCO’s, even in CMOS technology with on-chip low quality factor(Q) inductors.

5. SILICON TECHNOLOGIES

Explosive development of smaller and faster silicon transistors, driven by demand for faster digital processors, hascreated silicon technology with CMOS transistors suitable for RFIC’s for wireless applications. While 0.25µm silicontechnology is now a widely available, mature process, 0.18µm technology is also in volume production with ft valuesabove 50 GHz, and smaller features are still to come. CMOS technology used to be considered inferior to bipolar due tolower gm/I, and thus unsuitable for RF circuits. However, increased device speed has compensated for many drawbacksof these devices, and made it possible to explore its advantages, such as higher linearity and lower minimum NF. Novelcircuit design also enabled successful VCO design despite intrinsic high 1/f noise. As a result, CMOS LNA’s, mixersand VCO’s, competitive with performance of silicon bipolar circuits, have been realized.

Performance of silicon IC’s is limited both by active and passive devices. Passive devices typically outnumber activedevices in RFIC’s, and also occupy most of the chip area. While transistor size had been scaled down dramatically dueto the demands of digital integrated circuit technology, passive components have not improved much from thisdevelopment. Relatively low substrate resistivity of silicon, on the order of 10 Ω-cm, introduces parasitic couplingbetween adjacent circuit sections and limits the performance of passive devices. Limited thickness of interconnect metalsalso contributes to the loss of passive devices. Integrated inductors are most commonly realized in planar spiralconfiguration. Metal thickness, substrate loss, and physical layout determine the quality factor of these inductors. Lowquality factor (Q) can significantly degrade circuit performance, such as VCO’s phase noise when used in an on-chipresonator or LNA’s NF when used for input matching.

5.1 LNALow noise amplifier is typically the first stage in wireless receivers (Fig. 1), and thus determines the sensitivity of thereceiver chain. Linearity is also an important parameter, particularly in BTS applications, since a high dynamic range ofwanted and unwanted signals can be present at the receiver input. Therefore an LNA typically must exhibit low NF andhigh IIP3 simultaneously. To satisfy GSM and UMTS BTS requirements discussed in previous sections, NF should bebelow 2 dB, and IIP3 above 10 dBm. Since CMOS devices exhibit minimum NF of under 0.5 dB [12] and low third-order intermodulation distortion due to near square-law current versus voltage behavior [13], it should be possible to usethem for LNA BTS applications. It has already been demonstrated that CMOS LNA’s can achieve excellent noise figure.A 5 GHz CMOS LNA with on-chip input inductor exhibited NF of 2.5 dB [14], and a NF lower than 1 dB was achievedfor a CMOS LNA at 1.2 GHz [15]. It was also demonstrated that it is possible to achieve both low NF and high IIP3 forLNA’s in 0.25µm BiCMOS technology, using optimum device size and bias current [16]. However, one inherentdrawback of bipolar devices regardless of the technology is that bias current must be low for low NF, and high for highIIP3, and thus cannot be optimized simultaneously to satisfy both requirements. In the case of CMOS devices, goodlinearity and noise figure can be achieved for same bias conditions. Since CMOS devices exhibit lower third-orderintermodulation products that bipolar devices in BiCMOS technology [17], CMOS LNA’s designed with optimumdevice size and bias condtitions should be able to achieve good NF and IIP3 simultaneously.

While CMOS devices exhibit lower minimum NF than bipolar devices, their input impedance is largely capacitive andthus more difficult to match to 50 Ohms. An inductor is required at the device input to provide impedance matching.However, on-chip inductors with Q values of under 10 can increase the LNA NF by over 0.5 dB. Micromachined RFinductors, compatible with silicon processed can be used to overcome this problem. These techniques result either ininductors suspended in air [18], or on a thick low-loss dielectric [19], removing them from a lossy silicon substrate inboth cases. If such processes can be added to standard CMOS technology at low cost, they would enable commercial useof CMOS RFIC’s even for most demanding wireless applications such as 2G and 3G base station receivers.

5.2 MixerReceiver linearity is typically dominated by mixer performance. In RF radios, both active and passive mixes can be used.Active mixers provide conversion gain, and thus reduce noise contributed by subsequent stages. These mixers,commonly used in handsets, are usually realized in Gilbert cell configuration, in either bipolar or CMOS technology [8].However, active mixers usually have a high NF (above 10 dB), and low IIP3 (below 10 dBm). After system partition,and depending on the front-end configuration, base station linearity requirements discussed in previous sections result ina mixer specification for IIP3 of 15-20 dBm. Passive mixers achieve a higher linearity than active mixers, and are thus

more suitable for base station receivers. Due to the fact that a single MOSFET transistor device can be used as a switch,CMOS mixers can also be realized in passive form. Resistive FET mixer has very high input IP3 compared with othermixer types [20]. Traditionally, such a mixer has been implemented using GaAs FET devices. However, a resistivemixer using silicon MOSFET can achieve same performance with much lower cost. In addition, it requires no DC bias.The resistive CMOS mixer, designed for GSM900 and DCS1800 base stations, with a double balanced ring structure anda broadband performance is described in [21]. This mixer has conversion loss better than 6.5 dB at both bands, and itachieves very high input IP3 of 18.9 dBm and 21.9 dBm at 900 MHz and 1800 MHz, respectively. Drawback of passivemixers is that they require a higher LO drive than active mixers. The mixer described in [21] has demonstrated a gracefuldegradation in conversion loss of 1.3 dB or less over a falling LO power level of 14 dBm to 4 dBm.

5.3 VCOSilicon bipolar transistors have traditionally been preferred devices for VCO’s, due to their low 1/f noise. However,even with silicon BJT’s, fully integrated VCO phase noise performance is limited by low-quality factor of on-chipinductors [13]. Recent advances in understanding of phase noise theory have helped to partially overcome this problem.Fully integrated bipolar oscillator realized in 0.25µm BiCMOS technology, designed was GSM applications, wasreported in [5]. In this design, the transistor is biased in a nonlinear mode such that it is saturated at part of the cycle, inorder to reduce the phase noise. This increases the harmonic content of oscillation and produces a waveform that hassteeper slope, and thus lower phase sensitivity to noise, at zero crossing. Phase noise at 600 kHz, 800 kHz, 1.6 MHz,and 3 MHz offset frequencies was measured to be –129 dBc/Hz, -132 dBc/Hz, -142 dBc/Hz, and –148 dBc/Hz,respectively. This phase noise performance meets the DCS 1800 MS, GSM 900 small MS (handset with output powerless than 2W), and DCS 1800 Micro BTS requirements, and comes very close to meeting DCS 1800 normal BTSrequirement. Among integrated VCO’s with on-chip resonators in a silicon-based process, this remains the lowest phasenoise reported to date [5].

Even though CMOS devices exhibit higher 1/f noise than bipolar devices, it is still possible to realize good phase noiseCMOS VCO’s using advanced circuit design techniques. Phase noise as low as -152 dBc/Hz at 1 MHz offset has beenreported for 0.35µm CMOS VCO at 1 GHz, and –148 dBc/Hz at 15 MHz offset at 2 GHz [22]. Such an oscillator wouldbe adequate for UMTS base station receivers. CMOS VCO realized in 0.25 µm with oscillation frequency as high as 50GHz has been reported in [23], with phase noise of below –99dBc/Hz at 1 MHz offset.

6. BiCMOS GSM BASE STATION RECEIVERS

A GSM900 and DCS1800 base station receivers were sucessfully realized in silicon-MMIC 0.25 µm BiCMOS process.The receiver chips achieve low noise figure and high third order intercept (IP3) simultaneously without any gain control.The chips were fabricated using Agere (formerly Lucent Microelectronics) BiCMOS process [24] with inductor qualityfactor (Q) of approximately 10. With a supply voltage of 3 V, current consumption is 132 mA for the GSM900 receiverand 117 mA for the DCS1800 receiver. This was the first reported silicon-integrated radio front end for base stations [4].

The block diagram of the RF receiver is shown in Figure 5. The LNA before mixer is partitioned into LNA1 and LNA2.The input to LNA1 is connected to duplexer. A switch between LNA1 and LNA2 is needed for loop test once the chip ismounted in the system. The output of LNA2 goes to a 3 dB power divider of which one output is used for diversity whilethe other goes back on chip after a dielectric filter. The dielectric filter is used to reject image band noise and other out-of-band interference. The on-chip balun converts single-ended input to differential signal for the balanced mixer. Theon-chip LO buffer [25] converts single-ended LO input to differential output and amplifies its power level to drive thedouble balanced mixer [21]. IF port of the mixer goes to the IF buffer before a SAW filter or IF sampling. All of thebuilding blocks are integrated in a 0.25 µm silicon BiCMOS process except the 3 dB power divider and the dielectricfilter. All of the RF inductors required in the circuits are integrated on-chip.

Fig. 5 BiCMOS receiver block diagram.

The partition of gain, noise figure, and IP3 is optimized to achieve the best overall performance. LNA1 and LNA2 havesimilar structure except LNA1 is optimized for low noise figure and LNA2 is optimized for high IP3. This is done byoptimizing the bipolar transistor size, operating current, and degeneration inductance. The switch between LNA1 andLNA2 not only functions as a loop-test switch, but also has a variable attenuation control that can be used to adjustoverall receiver noise figure and linearity. The LNA1 achieves noise figure under 2dB at both bands, whereas LNA2achieves output IP3 greater than 22 dBm at both bands [16]. Bipolar transistors are used for LNA1, LNA2, and LObuffer whereas CMOS transistors are used for switch, mixer, and IF buffer. The chip layout is designed to fit into aTQFP-48 package with 7 mm x 7 mm body size. Both chips have a similar size of about 3.5 mm x 3.5 mm.

Fig. 6 DCS1800 receiver photograph. Chip size is 3.5 mm x 3.5 mm.

LNA1 SWITCHLNA2

LO BUFFERMIXER RF

BALUN

LO Buffer

LNA1LNA2Switch

Mixer

Balun Filter

3dB Divider

IF Buffer

Loop Test IN

RF IN

Diversity OUT

IF OUTLO IN

Integrated in Silicon

To reduce the ground inductance and improve the performance, the ExposedPad TQFP-48 package is used [26]. Thepackage has a backside ground pad that provides a very good RF ground to the chip. With a number of bond wiresdirectly bonded to the ground pad, the total ground inductance is reduced to below 0.5 nH. Since the chip size is limitedin this package, the CMOS IF buffer amplifier is packaged separately. This also allows the reuse of IF amplifiers afterchannel selection SAW filters to optimize overall system sensitivity. The die photo of DCS1800 integrated chipincluding LNA1, switch, LNA2, mixer, RF balun, and LO buffer is shown in Figure 6. The GSM900 integrated chip issimilar in layout to DCS1800 chip.

The integrated receivers achieve better than 2.1 dB noise figure, 25.3 dB gain, -1.9 dBm input IP3 in the GSM900 bandand better than 3.3 dB noise figure, 20.7 dB gain, and 0.4 dBm input IP3 in the DCS1800 band. Both receivers show aflat response over their respective bandwidth. A summary of each receiver performance, as well as individual circuits isshown in Figure 7. The building blocks were tested on-wafer and the integrated receivers were tested in packages. Thesechips were the first silicon RFIC’s developed for base station receivers. The measured results demonstrates thefeasibility of making compact and low-cost transceiver units for base station applications, which is essential to thehardware implementation of 3G wireless communication systems utilizing multiple antennas and front end radios toincrease capacity.

Fig. 7 GSM900 and DCS1800 receiver and individual circuit performance summary.

GSM900Receiver

NF(dB)

Gain(dB)

P1dB

(dBm)IIP3(dBm)

OIP3(dBm)

Current(mA)

LNA1 1.4 13.0 -5.5 9.3 22.3 12.9

LNA2 2.5 9.6 3.0 17.0 26.0 32.9

BroadbandSwitch 2.2 -2.2 17.5 30 27.8 0

Mixer w/LO Buffer 6.3 -6.3 10.3 18.9 12.6 45.8

IF Buffer 1.3 15.8 2.0 13.1 28.9 40

IntegratedReceiver 2.1 25.3 -8.2 -1.9 23.6 132

DCS1800Receiver

NF(dB)

Gain(dB)

P1dB

(dBm)IIP3(dBm)

OIP3(dBm)

Current(mA)

LNA1 1.9 11.3 -9.0 3.7 15.0 10.7

LNA2 2.1 12.2 -3.0 10.7 22.9 22.1

BroadbandSwitch 3.3 -3.3 18.5 27 23.7 0

Mixer w/LO Buffer 6.4 -6.4 11.3 21.9 15.5 44.6

IF Buffer 1.3 15.8 2.0 13.1 28.9 40

IntegratedReceiver 3.3 20.7 -9.0 0.4 21.6 117

7. CONCLUSIONS

While GSM BTS specifications present a challenge in terms of receiver front-end noise figure, linearity, and phasenoise, UMTS standard requires lower NF, but has more relaxed linearity and phase noise specifications. Even thoughthese specifications are still challenging to achieve in silicon MMIC technology, it is possible to meet them with theoptimum specification partition, device choice, and circuit configuration. Performance of a fully integrated BiCMOSradio receiver front-end described in this paper, demonstrates that it is feasible to meet GSM BTS specifications withlow-cost silicon technology. Since UMTS specifications result in radio requirements similar to those for 2G radios, 3Gbase station silicon RFIC’s should become available in the near future. Recent advances in CMOS technology wouldmake it feasible to meet these requirements at even lower cost.

ACKNOWLEDGMENTS

The authors would like to thank Lucent GSM group in Swindon, UK for collaboration, Agere (formerly LucentMicroelectronics) for chip fabrication, and R. Whitehouse from Agilent Technologies for providing testing equipment.Technical discussions with C. Zelley, V. Archer, Y-J. Chen, V. Lubecke, and E. Westerwick, and chip layout by F.Hrycenko and T. Gabara are greatly appreciated.

REFERENCES

1. GSM 05.05 version 5.5.1 (ETS 300 910): "Digital cellular telecommunications system (Phase 2+); Radiotransmission and reception," January 1998.

2. M. Steyaert at all, “A Single-Chip CMOS Transceiver for DCS-1800 Wireless Communications,” IEEE ISSCCDigest of Technical Papers, pp. 48-49, February 1998.

3. T. Cameron at all, “A Multi-Chip Receiver Module for PCS1900 and DCS1800 GSM Basestations,” 1999 IEEEWireless Comm. And Networking Conf. Digest, pp. 732-726.

4. J. Lin, O. Boric-Lubecke, P. Gould, C. Zelley, Yung-Jin Chen, Y. Ran-Hong "3V GSM Basestation receivers using0.25µm BiCMOS," IEEE ISSCC Digest of Technical Papers, pp. 416-471, February 2001.

5. J. Lin, “An Integrated Low-Phase-Noise Voltage Controlled Oscillator for Base Station Applications,” IEEE ISSCCDigest of Technical Papers, pp. 432-433, February 2000.

6. J. Jussila, J. Ryynanen, K. Kivekas, L. Sumanen, A. Parssinen, and K. Halonen, “A 22 mA 3.7 dB NF DirectConversion Receiver for 3G WCDMA, “IEEE ISSCC Digest of Technical Papers, pp. 284-285, February 2001.

7. GSM World, www.gsmworld.com8. L. E. Larson, RF and Microwave Circuit Design for Wireless Communications, Norwood, MA, pp. 365-369, 1997.9. B. Razavi, RF Microelectronics, Prentice Hall, 1998.10. 3rd Generation Partnership Project; Technical Specification Group Radio Access Networks; UTRA (BS) FDD;

Radio Transmission and Reception, 3GPP TS25.104, V4.1.0, 2000-06.11. H. Holma and A. Toskala, WCDMA for UMTS-Radio Access for Third Generation of Mobile Communications,

Revised Edition, John Wiley&Sons, 2001.12. T. H. Lee, “CMOS RF: (Still) no Longer an Oxymoron,” 1999 IEEE RFIC Symp. Digest, pp. 3-6, Anaheim, CA,

June 1999.13. L. E. Larson, “Integrated Circuit Technology Options for RFIC’s-Present Status and Future Drections,” IEEE

Journal of Solid State Circuits, vol. 33, no. 3, pp. 387-399, March 1998.14. E. H. Westerwick, “A 5-GHz Band CMOS Low Noise Amplifier with a 2.5-dB Noise Figure,” 2001 Int. Symp. On

VLSI Technology, Systems, and Applications, pp. 224-227, Hsinchu, Taiwan, April 2001.15. P. Leroux, J. Janssens, and M. Steyaert, “A 0.8 dB NF ESD-protected 9 mW CMOS LNA,” IEEE ISSCC Digest of

Technical Papers, pp. 410-411, February 2001.16. O. Boric-Lubecke, J. Lin, T. Ivanov, Y. H. Ran, "Si-MMIC BiCMOS low-noise, high linearity amplifiers for base

station applications," IEEE Asia-Pacific Microwave Conf. 2000, pp. 181-184, Sidney, Australia, December 2000.

17. N. Suematsu, M. Ono, S. Kubo, M. Uesugi, and K. Hasegawa, “Intermodulation Distortion of Low Noise SiliconBJT and MOSFET Fabricated in BiCMOS Process,” IEICE Trans. on Electronics, vol. E82-C, pp. 692-698, May1999.

18. P. L. Gammel, B. P. Barber, V. M. Lubecke, N. Belk, and M. R. Frei, “Design, Test and Simulation of Self-Assembled, Micromachined RF Inductors,” SPIE Symp. on Design, Test, and Microfabrication of MEMS andMOEMS, pp. 582-591, Paris, France, April 1999.

19. J. Rogers, V. Levenets, C. A. Pawlowicz, N. G. Tarr, T. J. Smy, and C. Plett, “Post-Processed Cu Inductors withApplication to a Completely Integrated 2-GHz VCO,” IEEE Trans. Elec. Devices, vol. 48. no. 6, pp. 1284-1287,June 2001.

20. S. A. Maas, Microwave Mixers, 2nd Edition, Artech House, 1993.21. P. Gould, C. Zelley, and J. Lin, “ CMOS Resistive Ring Mixer MMICs for GSM 900 and DCS 1800 Base Station

Applications,” IEEE MTT-S Intr. Microwave Symp. Proc., pp. 521-524, Boston, MA, June 2000.22. E. Hegazi, H. Sjoland, and A. Abidi, “A Filtering Technique to Lower Oscillator Phase Noise,” IEEE ISSCC Digest

of Technical Papers, pp. 364-365, February 2001.23. H. Wang, “A 50 GHz VCO in 0.25υm CMOS,” IEEE ISSCC Digest of Technical Papers, pp. 372-373, February

2001.24. Y-F Chyan at all, “A 50 GHz 0.25µm Implanted-Base High-Energy Implanted-Collector Complementary Modular

BiCMOS (HEICBiC) Technology for Low-Power Wireless-Communication VLSIs,” 1998 IEEE Bipolar/BiCMOSCircuits and Technology Meeting ’98, Conference Proceedings, pp. 128-131, Minneapolis, September, 1998.

25. J. Lin, C. Zelley, O. Boric-Lubecke, P. Gould, and R. Yan, "A silicon MMIC active balun/buffer amplifier withhigh linearity and low residual phase noise," IEEE MTT-S International Microwave Symposium Proceedings, pp.Boston, June 2000.

26. Amkor ExposedPad TQFP Technical Data Sheet.