Whitepaper. Integrated LNA Serves Base Station Needs. Chin-Leong Lim

Transcription

1 Integrated LNA Serves Base Station Needs Whitepaper Chin-Leong Lim As demand for wireless broadband grows, so does the importance of cellular base station (BTS) performance. The low noise amplifier (LNA) on the antenna tower is a key element in establishing BTS performance, affecting both the base station s area of coverage and its tolerance of other nearby transmitters. Integrated LNAs with active bias regulators, such as Avago Technology s MGA-63x family, provide the noise figure and linearity needed for outstanding BTS performance. A cellular base station (BTS) design should provide a wide area of coverage while sharing a tower site with multiple wireless transmitters. These attributes allow service providers to use as few tower sites as possible to serve a region and to share the cost of those sites with other users. To meet these coverage and sharing requirements, the BTS architecture needs receivers with high sensitivity to pull in distant cell phone signals while also rejecting strong in-band and out-of-band transmissions from other wireless transmitters nearby. Receiver sensitivity defines the weakest signal that a wireless receiver can recover and is the product of many factors, including receiver signal bandwidth (BW, in Hz) and the information rate the signal must support. This sensitivity can be described as: Rx_sen (dbm) = log BW + SNR + F (1) where SNR is the signal to noise ratio required to support the desired information rate, and F is the system noise figure. The Friiss equation, stated as: F 2 1 F 3 1 F = F (2) G 1 G 1 G 2 where G n is the gain of the nth stage in the receive chain, shows that the noise figure (F 1 ) of the first amplifying stage in a receiver chain dominates the system s total noise, while the noise performance of subsequent stages (i.e., F 2, F 3 and so on) are of lesser importance. Thus, a Low Noise Amplifier (LNA) in the first stage, as its name implies, improves receiver sensitivity by minimizing the cascade noise figure, F. Current cellular base-stations (BTS) typically locate the LNA near the antenna in the aerial tower. This positioning helps mitigate noise figure degradation that would otherwise result from loss in the cable coming down the tower from the antenna to a remote LNA stage. In the BTS architecture two additional elements, a transmit-receive (Tx-Rx) diplexer for duplexing a common antenna and an interference filter for preventing out-of-band blocking or desensitization, typically precede the LNA stage. Both duplexer and filter, however, have losses. Because these losses occur before amplification they need to be small to keep SNR as high as possible. An LNA with an extra margin in noise performance, though, will relax the duplexer-filter s loss requirement. The LNA in BTS designs does need to address other critical performance requirements beyond a low noise figure. These requirements may include high gain, to overcome loss in the long cable connecting a tower-mounted LNA to the ground-level radio cabin, as well as high linearity. The high linearity is needed to prevent distortion that could cause inter-channel interference when dealing with strong signals.

2 Low-noise techniques in amplifier design While high gain and high linearity are important, an LNA s noise figure is its defining and most important characteristic. There are many design and fabrication techniques that can affect an amplifier s noise figure (F), but some are not suitable for BTS applications. The Fukui equation, for instance, shows that the physical temperature of an electronic amplifier (T PHY ), directly affects its noise figure. Due to this relationship, lowering T PHY to near 0K using closedcycle helium cooling is by far the most effective noisereduction method available and has demonstrated F 0.05 db at 900 MHz. The maintenance challenges and high cost (~ USD10 k per cooler) of cryogenic refrigeration, however, make such cooling impractical for all but the most performance-critical applications such as radio telescopes and earth stations for interplanetary probes. Other factors that strongly influence an amplifier s F are more controllable. The choice of semiconductor material for transistor fabrication, for instance, strongly influences F. Cutting-edge materials, such as Indium Phosphide (InP), allow unrivalled noise performance. These, however, are usually too costly for routine commercial use. On the other hand, Silicon CMOS offers unbeatable cost benefits but offers only a modest noise level (Table 1). Gallium arsenide materials, especially enhanced high electron mobility transistor technology (phemt), offer a good balance of cost and performance. Table 1. Relative noise performance and cost comparison of LNA semiconductor technologies. Technology Noise rating Epitaxy cost/mm 2 InP HEMT Very good $10 GaAs phemt Very good $2 GaAs HBT Good $2 Si CMOS Fair $0.01 Packaging choices can also affect noise performance, specifically by attenuating high-frequency signals through dissipation of EM field energy in the material surrounding IC leads and internal wiring. In the 70s and 80s, low-noise microwave amplifiers typically used ceramic-packaged devices because ceramic has extremely low loss (dissipation factor, tan = 0.001). Further, ceramic packages supported the use of strip-line leads that could be width matched to the PCB trace to minimize discontinuities. The shift for cost-savings reasons in the 90s to plastic surface-mount packaging (SMP), such as a SOT-23 or SC-70, resulted in packaging that significantly degrades noise performance because of the epoxy encapsulation s higher loss (tan = to ). Also, the abrupt width changes at the die to bond-wire and lead to micro-strip interfaces increased reflection losses. At the device level, most RF parameters including noise can be improved by shrinking transistor feature size (i.e., gate length). It has been shown, for instance, that reducing the CMOS feature size from 0.18 m to 90 nm resulted in a useful 0.2 db noise improvement at ~1 GHz. The drawback has been a significantly higher manufacturing cost. In addition to these device-level techniques there are circuit level techniques for reducing noise. In amplifier design, for instance, there is often a need for impedance matching. But there can be a significant difference between the input conjugate match ( S ) that ensures maximum source signal transfer to the amplifier and the optimum noise match ( opt ) that minimizes noise figure. This difference typically requires sacrifice of the amplifier s noise performance during matching in order to avoid an input return loss (IRL) that increases with the generatorto-input impedance transformation ratio. The IRL could be minimized by increasing the amplifier s unloaded Q via a cavity or silver-plated resonators, but these esoteric components are either too cumbersome or costly for massproduced commercial products. It has been shown, though, that adding a small inductance (L S ) in the source to ground path can lower the S - opt divergence to reduce the sacrifices that arise from favouring the input conjugate match. Researchers have reported as much as a 0.15 db noise reduction at 1.95 GHz when using this technique. In practice, however, only a small and lessthan-ideal amount of L S can be added before undesirable peaks in the frequency response begin to form far above the design pass-band. Another noise-reduction circuit technique utilizes parallelism to address the matching compromise. FET-type devices have relatively high optimum noise impedance (Z opt ), and so connecting two or more such transistors in parallel can lower noise by reducing the mismatch between Z opt and the generator impedance (Z s ). Designs reported to use this method include a 3 x FET in the MHz VHF FM broadcast range and 2 x HEMT at 1.4 GHz. 2

3 Avago MGA-63x device characteristics In creating the MGA-63x line of microwave monolithic integrated circuit (MMIC) LNA devices, Avago Technologies employed the most useful of these and other noisereduction techniques. The device packaging, for instance, employs a compact (2 x 2 x 0.75 mm) 8-pin quad flat non-lead (QFN) structure that eliminates many of the reflection-causing discontinuities inherent in wire-bonded lead-frame packages. Each member of the device family utilizes the same pin out and footprint, allowing developers to create a single circuit board design that can handle multiple frequency bands ranging from 400 MHz to 4 GHz through simple component substitution. The MGA-63x family s fabrication technology is a proprietary 0.25 m feature-size GaAs enhancement-mode, pseudo-morphic high electron mobility transistor (ephemt) process that has a high gain-product bandwidth (f > 30 GHz). This value allows the LNA s target gain (> 17 db at 0.9 GHz) to be achieved in one transistor stage. The process minimizes Johnson noise generated in the interconnections by making the metal twice as thick as in previous process iterations. In addition, use of high-conductivity metallization helps keep the packaged device s noise figure comparable to that of ceramic devices. The fabrication technology helps the MGA-63x family resist blocking, which desensitizes a receiver by lowering the gain and increasing F in the presence of strong signals. A non-synchronous interferer, such as a powerful transmitter sharing the same tower, or a synchronous source, such as a transmission that leaks past the circulator or duplexer in a transceiver with simultaneous transmit and receive capability, can cause blocking. Components with a high gain compression threshold resist blockers. Gain compression is primarily caused by non-linear transfer characteristic in an amplifier that is driven beyond the Vbias [1] RFin [2] NU [3] NU [4] ESD PROTECT AMP ACTIVE BIAS L S [8] NU [7] RFout [6] NU [5] NU Figure 1. The MGA-63x family integrates a high-gain transistor and active bias regulator to form an integrated LNA with outstanding linearity and a low noise figure. linear region with increasing heat dissipation as a minor contributor. The low knee voltage (0.3 V) of the process permits a large voltage swing before clipping, giving the device a high gain compression threshold. In addition, the GaAs substrate s comparatively low bulk conductivity helps minimize heat loss. The MMIC itself consists of a single-fet common-source amplifier and an active bias regulator (Figure 1). The active bias helps improve linearity in the LNA s operation. A small source inductance (L S ) enables good IRL and low F to be simultaneously achieved at one S value. (For full specification details please refer to the data sheet.) Because the regulator and the LNA transistors are integrated they have undergone similar processing, which allows V bias and V GS to mirror each other. This mirroring helps ensure I ds temperature stability by correcting for any thermal drift. It also helps compensate for variations in transconductance between wafer runs. The result is highly consistent field operation by the MGA-63x family. The bias regulator allows adjustment of the LNA s quiescent current (Ids) either through an externally applied voltage, V bias, or by connecting a resistor from V dd to the bias input. The regulator s low current drive requirement (I bias 1 ma) is compatible with most CMOS families and makes it possible for a microcontroller to directly switch the LNA on and off in time-domain multiplexed (TDM) applications The adjustable bias feature provides a convenient mean to trade-off linearity for power consumption. In applications that do not require highest linearity, designers can choose to conserve power by using an R bias value that is larger than nominal (6.8 k ). Alternatively, the LNA s OIP3 can be varied as much as 10 db by changing Idd over the 25 to 75 ma range, with minimal effect on gain and output power match ( G and P1dB 0.5 db). This opens the possibility of designing an LNA that can adaptively respond to degree of spectrum crowding through microcontroller regulation of V bias. Both the transistor design and the bias regulator work to avoid requiring an external matching network that would increase insertion loss and hence reduce F in the LNA. The transistor device geometry is dimensioned and its nominal bias current set to make its input impedance close to 50. The integrated active bias regulator circuit avoids affecting the LNA s input impedance the way an external resistor bias circuit might. Together these features eliminate the input match requirement, helping minimize the LNA s noise figure. 3

4 Evaluation circuit proves performance To demonstrate the MGA-63x family s exceptional performance, Avago created and tested a cellular basestation LNA for operation at 900 MHz (Figure 2) using an MGA-633P8 MMIC and a minimal number of passive components for matching and biasing which are not feasible for integration at the chip level. The component values were chosen based on a single design simulation cycle, not optimized by bench tuning. (For full discussion of the evaluation board simulation and development, see Avago application note AN-5457, MGA-633P8 GaAs MMIC LNA Enables 900 MHz BTS Amplifier with Industry Best Noise Figure and Linearity.) Even so, the evaluation board demonstrated several key features of LNAs designed using the MGA-63x family. For one, the demonstration board showed that because the input inductor (L 1 ) essentially functions as an RF choke only, the amplifier s input loss is relatively insensitive to input resonator s unloaded Q value (Q UL ). Simulation shows that F < 0.05 for any Q UL in the 20 to100 range. This range represents using a 0402-size multilayer chip inductor at the lower end and a much larger air-core inductor at the upper end for the amplifier s input inductor. This uncritical input resonator requirement enables low-cost and compact LNA designs. The printed circuit board (Figure 3) of the evaluation board reflects demonstrates how compact a design can be. The board measures 21.5 x 18 x 1.4 mm and comprises microstrip with co-planar ground on a 10 mils Rogers RO4350 a mid-priced material with modest RF performance that is FR4-process compatible. A lower-cost FR4 material of 1.2 mm thickness bonded to the RO4350 ground-plane provides rigidity. RF connections were made through edge-launch SMA-to-micro-strip transitions (Johnson Component P/N ), while the DC supply was connected via a 2-pin straight PCB header. Because of the non-critical input resonator requirement, 0402-size chips could be used to shrink the area populated by components to ~ 8 x 10 mm 2. At 900 MHz, the demonstration LNA achieved F 0.3 db, gain (G) 18 db, and both input and output return loss (IRL and ORL) better than -15 db (Figure 4). The stability of these return losses over a 1 GHz span shows a wide bandwidth for the input and output match, which is very favourable from the system standpoint. Traditionally, BTS LNAs rely on either an isolator or a quadrature hybrid coupler (balanced LNA) to achieve the desired input match. This design s low IRL allows the high-loss and costly isolator/quadrature coupler to be eliminated in many applications. Further, the very wide bandwidth of the input and output match prevents detuning of a BTS s diplexer and input/output filters by termination reflections, thus lowering the complexity requirements (and cost) of these system elements. in V BIAS C3 C1 L1 R1 C1, C2 100 pf Murata GRM15 L1, L2 33 nh Toko LL1005 C4 27 pf Murata GRM15 C3, C5 4.7 F Murata GRM15 R1 9.1 Ohm Q1 MGA633P8 Avago Note: All SMD components are 0402 size [1] [2] [3] [4] R BIAS [8] [7] [6] [5] L2 Vdd Figure 2. This demonstration circuit shows a complete 900 MHz BTS LNA design for evaluation of the Avago MGA-63x family x 10 Figure 3. The entire LNA occupies a PCB not much larger than the connectors needed for the RF I/O. C2 C5 C4 out 4

5 S11 mes db S11 sim IRL, ORL & G vs. Frequency S22 mes S21 mes S22 sim S21 sim -25 Centre: MHz #2, 5 V 53 ma Span: 1.00 GHz Figure 4. Gain and return loss plots show the frequency insensitivity of the LNA s matching networks. Although the main target market for the MGA-63x family is narrowband cellular BST, the RF performances (F < 0.4 db and R < -10 db in the 400MHz ~ 1400 MHz range) demonstrated by the evaluation board show that the devices are also suitable for many wideband/multi-bands applications such as cable\satellite TV distribution infrastructure, scanners, military applications, and multi-service radios. The evaluation board results confirm that the MGA-63x active-bias, ultra-low-noise amplifier family provides base station designers with numerous essential and valuable features. In addition to providing the high linearity and low noise figure needed for cellular communications, these devices offer high gain for single-stage LNA designs, enough gain margin to ease design demands on duplexer-filters in the signal path, active bias control for adaptive power/linearity control, and inherent 50 input impedance to eliminate the need for matching networks. The result is simpler, smaller, and more capable LNA designs at lower cost. For product information and a complete list of distributors, please go to our web site: Avago, Avago Technologies, and the A logo are trademarks of Avago Technologies in the United States and other countries. Data subject to change. Copyright Avago Technologies. All rights reserved. AV EN - September 30, 2011