2008:230 CIV MASTER'S THESIS Low-Noise Amplifier Design and Optimization Marcus Edwall Luleå University of Technology MSc Programmes in Engineering Electrical Engineering Department of Computer Science and Electrical Engineering Division of EISLAB 2008:230 CIV - ISSN: 1402-1617 - ISRN: LTU-EX--08/230--SE A BSTRACT Low-Noise Amplifiers are key components in the receiving end of nearly every communications system. The wanted input signal of these systems is usually very weak and the primary purpose of the LNA is consequently to amplify the signal while at the same time adding as little additional noise as possible. Its performance is measured in a number of figures of merit among which gain and noise figure are most notable while dynamic range, return loss and stability are examples of others. In May 2005 a four year design study entitled EISCAT 3D was initialized. Its purpose was to investigate the feasibility of a next-generation incoherent scatter radar system. One of the responsibilities of EISLAB at Luleå University of Technology is to design a receiver front-end, which include an LNA with extremely high performance requirements. For that reason a MATLAB Particle Swarm Optimization implementation was developed to iteratively find a solution to optimal component values for a user definable LNA topology. In this master’s thesis, the radio frequency concepts essential to traditional LNA design as well as the design procedure itself are explained. A description to the optimizer is then given, including a chapter on 2-port noise calculations. With the objective to find an LNA design with even higher performance than the previously designed EISCAT 3D LNA, four topologies are evaluated using the optimizer while consistently targeting the EISCAT 3D specifications. These topologies include the original reference design and one that employs the inductive source degeneration design technique. The latter showed significantly improved performance with an approximate 2 dB gain increase and 0.1 dB noise figure reduction while still maintaining the return loss and stability requirements. iii P REFACE This master’s thesis was carried out at the Embedded Internet Systems Laboratory (EISLAB), Department of Computer Science and Electrical Engineering, Luleå University of Technology and was descendant to the EISCAT 3D design study. Among the people that I would like to thank for their contribution to my master’s thesis are Dr. Jonny Johansson for his encouragement and insightful help, and my supervisor, Ph.D. student Johan Borg for his invaluable expertise. On a personal level, I wish to express gratitude to family and friends who have supported me throughout this journey. Marcus Edwall v C ONTENTS Chapter 1: Introduction 1 Chapter 2: Radio Frequency Concepts 2.1 Reflection . . . . . . . . . . . . . . . 2.2 Scattering Parameters . . . . . . . . 2.3 The Smith Chart . . . . . . . . . . . 2.4 The Quality Factor . . . . . . . . . . 2.5 Impedance Transformation . . . . . . . . . . . 3 3 5 6 7 7 . . . . . . . 9 9 14 15 15 17 17 19 Chapter 4: 2-ports and Noise 4.1 Method of Linear 2-port Noise Analysis . . . . . . . . . . . . . . . . . . . 23 23 Chapter 5: The Optimizer 5.1 Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Notable functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 27 28 28 Chapter 6: LNA Topology Evaluation 6.1 Circuit 1: Original . . . . . . . . . . . . 6.2 Circuit 2: Simplified . . . . . . . . . . . 6.3 Circuit 3: Without Feedback . . . . . . . 6.4 Circuit 4: Inductive Source Degeneration 31 32 32 33 34 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 3: Low-Noise Amplifier Design Strategy 3.1 Target Specifications . . . . . . . . . . . . . . . . 3.2 Active Device Selection . . . . . . . . . . . . . . . 3.3 DC Bias Network Design . . . . . . . . . . . . . . 3.4 Matching Network Design . . . . . . . . . . . . . 3.5 Noise Optimization . . . . . . . . . . . . . . . . . 3.6 Low-Noise Amplifier Topologies . . . . . . . . . . 3.7 Determining ΓS and ΓL . . . . . . . . . . . . . . . Chapter 7: Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 C HAPTER 1 Introduction The European Incoherent Scatter Association (EISCAT) is an international research organization that operates four incoherent scatter radars located in northern Scandinavia. These are used to provide ionospheric radar observations for geophysical environmental monitoring, modelling and forecasting as well as plasma physics research. To meet increasing demands and maintain world leadership in this field a four year design study, EISCAT 3D [1], is underway since May 2005. This next-generation incoherent scatter radar system is based on large phased array VHF antennas. The potential of the system will exceed all other similar facilities both existing and under construction. Among the involved parties is Luleå University of Technology (LTU) whose main responsibility is the phased array receivers, including a low-noise amplifier (LNA). Because of the extreme performance requisites, no commercially available pre-built product exhibit specifications near those required. Particular emphasis has therefore been put on design and optimization of these critical components. The work presented in this master’s thesis includes a description of conventional design and optimization techniques for an LNA. In addition, a MATLAB particle swarm optimization implementation, a tool developed within the EISCAT 3D framework, has been used to evaluate variations of an LNA design and optimize its associated performance parameters. A significant portion of the work was founded in firstly the recognition of underlying principles of LNA design and secondly understanding of the optimizer. 1 C HAPTER 2 Radio Frequency Concepts For a fundamental understanding of the Low-Noise Amplifier design procedure, it is necessary to introduce a series of underlying concepts. Covered topics include reflection, scattering parameters, the Smith Chart, the quality factor, and impedance transformation. Further information on this topic can be found in [2] and [3] on which this and the next chapter are based. 2.1 Reflection When a power wave travels through an impedance discontinuity, at that junction (Figure 2.1), a fraction of the wave will be reflected. As a consequence, the counterpart (the incident wave) will lose some of its magnitude. Naturally, this is an undesirable phenomenon in any application where power conservation is critical. The extent of incident power loss is related to the similarity of the impedances as seen in both directions from the junction. So the objective, in order to maximize the power transfer, is to optimize the impedance match. Further information on that subject follows in Chapter 3. There are a number of performance parameters that show to what extent the impedances are matched. Firstly, the Reflection Coefficient which by definition is the ratio of the reflected wave to the incident wave (Equation 2.1), but can also be expressed in terms of impedances. It is a complex entity that describes not only the magnitude of the reflection, but also the phase shift. ZL − ZS Ref lected wave = (2.1) Incident wave ZL + ZS Note that this is the load reflection coefficient with respect to the source impedance. It is also commonly expressed with respect to the characteristic impedance (Z0 ). When the load is short-circuited, maximum negative reflection occurs and the reflection coefficient assumes minus unity. In contrast, when the load is open-circuited, maximum positive ΓL = 3 4 RF Concepts ZS ΓL ZL Figure 2.1: Simple circuit showing the impedance discontinuity junction and measurement location of ΓL . Figure 2.2: Incident wave (solid), reflected wave (dashed) and standing wave (dotted). reflection occurs and the reflection coefficient assumes plus unity. In the ideal case, when ZL is perfectly matched to ZS , there is no reflection and the reflection coefficient is consequently zero. A closely related parameter is the Voltage Standing Wave Ratio (VSWR), which is commonly talked about in transmission line applications. As the incident and reflected wave travel in opposite directions the addition of the two generates a standing wave, see Figure 2.2. The VSWR is defined as the ratio of the maximum voltage to the adjacent minimum voltage of that standing wave (Equation 2.2). Knowing the domain of the reflection coefficient, it follows that when there is no reflection as in a perfectly matched system; VSWR assumes its minimum and ideal value of 1.0:1. V SW R = |V |max 1 + |ΓL | = |V |min 1 − |ΓL | (2.2) The Return Loss (RL) is simply the magnitude of the reflection coefficient in decibels 2.2. Scattering Parameters a1 b1 5 2-port a2 b2 Figure 2.3: A 2-port with incident waves a1 and a2 , and reflected waves b1 and b2 . (Equation 2.3). At times it is specified whether the return loss is measured on the inputor output side of the Device Under Test (DUT), with corresponding IRL and ORL naming. It should be mentioned that the return loss is occasionally expressed without the leading minus sign ending up with a negative RL. This is incorrect however as RL should be positive. RL = −20log|Γ| 2.2 (2.3) Scattering Parameters Scattering Parameters or S-parameters are complex numbers that exhibit how voltage waves propagate in the radio-frequency (RF) environment. In matrix form they characterize the complete RF behaviour of a network. At this point it is necessary to introduce the concept of 2-ports. It is fundamental in RF circuit analysis and simulation as it enables representation of networks by a single device. As the properties of the individual components and those of the physical structure of the circuit are effectively taken out of the equation, circuit analysis is greatly simplified. The characteristics of the 2-port is represented by a set of four S-parameters: S11 , S12 , S21 and S22 , which correspond to input reflection coefficient, reverse gain coefficient, forward gain coefficient and output reflection coefficient respectively. The concept of 2-ports is further described in Chapter 4 where noise calculations of linear 2-ports are addressed. There are alternative descriptive parameters for 2-ports, such as impedance parameters, admittance parameters, chain parameters and hybrid parameters. These are all measured on the basis of short- and open circuit tests which are hard to carry out accurately at high frequencies. S-parameters, on the other hand, are measured under matched and mismatched conditions. This is why S-parameters are favoured in microwave applications. S-parameters are both frequency- and system impedance dependent so although manufacturers typically supply S-parameter data with their devices it is not always ap- 6 RF Concepts plicable. Under such circumstances, it becomes necessary to measure the parameters. Referring to Figure 2.3, these measurements are carried out by measuring wave ratios while systematically altering the termination to cancel either forward gain or reverse gain according to the following equations: b1 S11 = a1 a2 =0 b1 S12 = a2 a1 =0 b2 S21 = a1 a2 =0 b2 . S22 = a2 a1 =0 Conclusively, the S-parameters relate the four waves in the following fashion: 2.3 (2.4) (2.5) (2.6) (2.7) b1 = S11 a1 + S12 a2 (2.8) b2 = S21 a1 + S22 a2 . (2.9) The Smith Chart The Smith Chart is a classic tool in RF engineering that has many uses and remains widely used although computers have become a convenient alternative. It is a fundamental aid in impedance matching network design and also serves as a standard for graphical presentation of impedance, reflectance, stability circles, gain circles, noise circles etc. In its most common form, the chart is made up out of two overlaid grids: the constant resistance circles and the constant reactance circles. The Cartesian coordinate system within the Smith Chart is used to plot the reflection coefficient. Furthermore there are three varieties of the Smith Chart: with impedance grid (Z Smith Chart), with admittance grid (Y Smith Chart) and the two combined (ZY Smith Chart). As the radius of the chart is unity, it is implied that all plotted values, whether they are impedances or admittances, must be normalized with respect to a reference (Equations 2.10 and 2.11). This reference is usually the characteristic impedance of the system which usually is 50 Ω. z= Z Z0 (2.10) y= Y Y0 (2.11) 2.4. The Quality Factor 2.4 7 The Quality Factor The Quality Factor (Q) is a descriptive parameter of the rate of energy loss in complete RLC networks or simply in individual inductors or capacitors. For the latter, Q is a measurement of how lossy the component is, that is how much parasitic resistance there is. So it follows that in applications where loss is undesirable, high Q components are advantageous. Additionally the Q factor is directly related to the bandwidth, where higher Q corresponds to narrower bandwidth. The equations for calculating Q are: QRLC = ω (2.13) XL ωL = R R (2.14) 1 |XC | = . R ωCR (2.15) QL = 2.5 (2.12) ω0 QRLC BW = QC = Etot Pavg Impedance Transformation As previously stated, in order to maximize power transfer from source to load, matching impedances is required. Specifically, in a circuit as seen in Figure 2.4 where the sourceand load impedances are fixed, the objective is to design the input matching network so that ZS matches Z1 and the output matching network so that ZL matches Z2 . In other words Z1 and Z2 respectively, are transformed to perceptually match the input and output impedances of the transistor. According to the Maximum Power Theorem, the maximum power transfer will occur when the reactive components of the impedances cancel each other, that is when they are complex conjugates. This is suitably called conjugate matching. To achieve the conversion with an impedance matching network of passive components, there are primarily three options. Firstly, there is the L-match. Its advantage is the simplicity, but that is simultaneously its downside as well because it has only two degrees of freedom. Since there are only two component values to set, the L-match is restricted to determining only two out of the three associated parameters: impedance transformation ratio, centre frequency and Q. To acquire a third degree of freedom, it is therefore desired to cascade another L-match stage. By doing so, another two types of impedance transformation matches are encountered: the π-match and the T-match (Figure 2.5). The advantages with the T- and π-match configurations do not end with an additional degree of freedom. But because of their topology they can absorb parasitic reactance present in source or load. Specifically the T-match will absorb parasitic inductance 8 RF Concepts Z1 Input matching network Output matching network Transistor ZS Z2 ZL Figure 2.4: Matching networks in a microwave amplifier. Zπ2 Zπ1 ZT1 Zπ3 ZT3 ZT2 Figure 2.5: Cascaded π- and T-matching networks. whereas the π-match will absorb parasitic capacitance. In addition it is also possible to achieve significantly higher Q compared to an L-match configuration. Another noteworthy impedance transformation option is bandpass filtering where the port impedances are unequal. C HAPTER 3 Low-Noise Amplifier Design Strategy In a system with a series of cascaded devices, where each stage adds additional noise that is potentially amplified along the way, it becomes evident that the very first stage and its noise and gain characteristics are critical. This is particularly true if the input signal is weak and has relatively large amounts of noise added to it. Hence, under these conditions, a Low-Noise Amplifier is applicable. It is, as the name suggests, an amplifier where particular emphasis has been put on its noise characteristics. Ftot = F1 + F2 − 1 F3 − 1 F4 − 1 + + + ... G1 G1 G2 G1 G2 G3 (3.1) Friis’s formula (Equation 3.1), with which the total noise factor of a system with cascaded stages is calculated, shows how F1 and G1 , the Noise Factor and gain of the first stage, dominate the overall Noise Factor. And so, it could be explicitly specified that the function of the Low-Noise Amplifier is to supply sufficient signal gain to overcome the noise of the succeeding stages while at the same time producing as little noise as possible itself. 3.1 Target Specifications As the first order of business when designing an LNA it seems appropriate to establish what the target specifications are. This is done in terms of a number of various parameters. 9 10 3.1.1 LNA Design Strategy Gain The gain of the device is its ability to amplify the amplitude or the power of the input signal. It is defined as the ratio of the output- to the input signal and is often referred to in terms of decibels (Equation 3.2). V oltage Gain = 2 Vout Rout 10log( V 2 in Rin ) = 20log( Vout ) Vin (3.2) Power gain is generally defined as the ratio of the power actually delivered to the load to the power actually delivered by the source. However, as simple as that may seem, this definition is not entirely relevant and is difficult to quantify since the source impedance in turn is difficult to specify. For that reason, a number of specific and therefore more useful definitions have evolved. Most notable are perhaps Transducer Gain- the ratio of average power delivered to the load to maximum available average power from the source, Available Power Gain- the ratio of maximum available average power at the load to maximum available average power from the source. As previously discussed, maximum power is only obtained when an amplifier is has complex conjugate terminations. 3.1.2 Noise Performance The fundamental noise performance parameter is the Noise Factor (F), which is defined as the ratio of the total output noise power to the output noise due to input source. If the Noise Factor is expressed in decibels it is called the Noise Figure (NF) (Equation 3.3). Another related and often talked about parameter in RF applications is the Signal-to-Noise Ratio (SNR), which is the ratio of the signal power and the noise power (Equation 3.4). The Noise Factor is equivalent to the ratio of the SNR at the input and that at the output of the LNA (Equation 3.5). Hence, the Noise Factor is a measure of to what extent the LNA degrades the SNR. An alternative way to express just that is the Noise Temperature (TN ), which is particularly useful with cascaded amplifier systems or in applications where the Noise Figure is extremely low, as it allows greater resolution. The Noise Temperature is calculated with a reference temperature (Tref ), that is normally 290 K. By definition TN is the temperature increase that is required in the source resistance, so that it alone produces the noise that corresponds to the output noise at Tref . Consequently, if there is no additional noise at the output of the amplifier, then TN is 0 K. N F = 10log(F ) SN R = Psignal Pnoise (3.3) (3.4) 3.1. Target Specifications 11 IP3 Pout 1dB Pin Figure 3.1: 1st-order output (solid), 3rd-order IM product (dotted). F = 3.1.3 SN Rin TN =1+ SN Rout Tref (3.5) Linearity The linearity of the LNA is another concern that must be taken into account. Linear operation is crucial, particularly when the input signal is weak with a strong interfering signal in close proximity. This is because in such a scenario there is a possibility for undesired intermodulation distortion such as blocking and crossmodulation. Third-order intercept (IP3) and 1-dB compression point (P1dB ) are two measures of linearity. IP3 shows at what power level the third-order intermodulation product is equal to the power of the first-order output. IIP3 and OIP3 are the input power and output power respectively, that corresponds to IP3. P1dB shows at what power level the output power drops 1 dB, as a consequence of non-linearities, relative the theoretical linear power gain, Figure 3.1. By knowing either IP3 or P1dB the other can be estimated with the following rule-of-thumb formula: IP 3 = P1dB + 10dB. (3.6) Both measurements indicate an upper distortion limit for the tolerable input power, whereas the noise figure sets a lower limit. The ratio of the two determines the dynamic range of the amplifier. Another similar measurement is the Spurious-Free Dynamic Range (SFDR), which in the LNA context usually relates to the greatest possible differential between the output signal power and the power of the third-order intermodulation product. This occurs at the point where the latter emerges above the noise floor, Figure 3.2. 12 LNA Design Strategy Pout C A B Pin Pnoise Figure 3.2: 1st-order output (solid), 3rd-order IM product (dotted). P1dB (A), IIP3 (B) and SFDR (C). 3.1.4 Stability In a stability perspective, an LNA can be either unconditionally stable or potentially unstable. Given the former condition, the LNA will not oscillate regardless of what passive source- and load impedance it is connected to. In a 2-port network, as seen in Figure 3.3, oscillation may occur when some load and source termination cause the input- and output impedance to have a negative real part. There are three main causes for this scenario: internal feedback, external feedback and excessive gain at out-of-band frequencies. To prevent instability, the aim is to place ΓS and ΓL in the stable region of the Smith Chart. In practice, this is done with filtering and resistive loading to attenuate gain. The condition for unconditional stability, in terms of S-parameters is K= 1 − |S11 |2 − |S22 |2 + |∆|2 > 1, 2|S12 S21 | (3.7) where ∆ = S11 S22 − S12 S21 . (3.8) It is common practice to graphically present the region for which the LNA is unconditionally stable with stability circles in the Smith Chart. That include the input stability circle in the ΓS -plane and the output stability circle in the ΓL -plane. The circumstances determine whether the stable region is inside or outside the stability circle according to Figures 3.4 and 3.5 3.1. Target Specifications ΓS 13 ΓOUT ΓIN ZS ΓL ZL 2-port ZOUT ZIN Figure 3.3: Stability of 2-port networks. rs rs CS CS Figure 3.4: Smith chart illustrating grey stable region in the ΓL plane. Left: |S11 | < 1, right: |S11 | > 1. CL rL CL rL Figure 3.5: Smith chart illustrating grey stable region in the ΓS plane. Left: |S22 | < 1, right: |S22 | > 1. 14 LNA Design Strategy -3dB ∆f f0 Figure 3.6: Illustration of centre frequency and bandwidth. 3.1.5 Centre Frequency and Bandwidth As the LNA will operate with input signals of a particular frequency band, it is desired to design it with a centre frequency and bandwidth accordingly. Looking at the transfer function of the LNA, the differential of the two points around the centre frequency f0 , where the power gain is halved, is the bandwidth, denoted ∆f in Figure 3.6. Although the target bandwidth should be specified numerically, by naming convention there are two options: narrowband and wideband. 3.1.6 Return Loss The Return Loss is a measure of how well the input impedance is matched to the reference impedance or how well the output impedance is matched to the load impedance in a power transfer perspective. Strictly speaking it signifies how much power is reflected due to impedance mismatch relative the transmitted power. Return loss is typically specified as IRL, which corresponds to the return loss at the input port. 3.2 Active Device Selection The properties of the active device are in many ways the final limiting factor for many parameters of the LNA. It is therefore good practice to select an active device with parameters (in terms of noise figure, gain and linearity) that correspond to and preferably exceed those of the target specifications. There are various active devices that are well suited for LNA applications including, but not limited to, the Heterojunction Bipolar Transistor (HBT), the Metal Epitaxial Semiconductor Field Effect Transistor (MESFET), the Modulation-Doped Field Effect Transistor (MODFET) and the High Electron Mobility Transistor (HEMT). In addition there are a multitude of varied semiconductor compounds that further extend the op- 3.3. DC Bias Network Design 15 Figure 3.7: A passive bias network [4]. tions. In the interest of limiting the scope of this thesis however, the device of choice is the GaAs Enhancement Mode Pseudomorphic High Electron Mobility Transistor (EPHEMT): Avago Technologies ATF-541M4 [4]. 3.3 DC Bias Network Design The purpose of the bias network is to set the quiescent point. That is the Vgs and Ids for a FET that causes it to operate in the preferred region. In a general perspective there are several types of biasing networks, although in LNA applications low complexity is desired and often sufficient. Typical passive and active bias networks can be seen in Figures 3.7 and 3.8. 3.4 Matching Network Design The concepts of impedance matching and impedance transformation have already been described in Chapter 2. To display how impedance matching is carried out in practice, it will be illustrated with an example [5] using the Smith Chart. Observe however, that there are other means, such as with computer software. Consider the circuit in Figure 3.9. The objective here is to maximize power gain, and therefore transform ZL into ZS∗ . Assume that the matching network topology is yet unknown. What is known however is that that ZL > ZS∗ , for that reason a downward impedance transformer L-match is used. Should a single L-match not be sufficient, additional shunt and series impedances 16 LNA Design Strategy Figure 3.8: An active bias network [4]. L ZS ZS* C ZL Figure 3.9: Matching network design example circuit. ZS =25-j15 Ω and ZL =100-j15 Ω. can be added incrementally until the target impedance is reached. Graphically, as it is be shown in the Smith Chart in Figure 3.10, the objective is to find a route to link the points corresponding to the normalized ZL and ZS∗ . First normalize ZL and ZS∗ with the system impedance. If it is unknown, use an arbitrary value in the same range as those of the load and source. In this example Z0 = 50 Ω is appropriate. zS∗ ZS∗ = 0.5 + j0.3 = Z0 (3.9) 3.5. Noise Optimization 17 ZL = 2 + j0.5 (3.10) Z0 Next, mark zS∗ and zL in the Smith Chart, then consider the first component, the shunt capacitor. Because it is a shunt component it is preferable to convert to admittance. Hence, find the point denoted A by rotating 180 degrees from zL . Next, go clockwise along the constant conductance circle to find point B. Because the value of the shunt capacitor is unknown, the length of the arc from A to B is also unknown. In this example however, as the arc for the series inductor has to be on the same constant resistance circle as zS∗ it is possible to geometrically find B. Finally, since the next component is in series, convert back to impedance to point D and from there go clockwise to zS∗ . The length of the arc A through B (b=0.78) is the normalized susceptance of C, whereas the length of the arc D through zS∗ (x=1.2) is the normalized reactance of L. To finalize, their respective values are calculated for a specified operating frequency of 60 MHz: zL = B = 41.4pF 2πf (3.11) X = 159.9nH. 2πf (3.12) B = bY0 = ωC ⇒ C = X = xZ0 = ωL ⇒ L = 3.5 Noise Optimization Since the match for optimum noise normally does not coincide with the optimum gain, linearity- or input match the objective is usually to find a satisfactory compromise. However, firstly to optimize the LNA exclusively for low noise, consider the general expression for noise in an LNA 2-port: Rn [(Gs − Gopt )2 − (Bs − Bopt )2 ]. (3.13) Gs Where Fmin is the minimum noise factor, Rn is the noise resistance, Gs is the source conductance, Bs is the source susceptance, Gopt is the optimum conductance and Bopt is the optimum susceptance. It becomes evident that for minimum noise the source admittance or impedance equivalent must appear to the LNA as Gopt + jBopt . To obtain a simultaneous noise and input match, Γ∗IN (that yields best gain and input match) has to be shifted graphically in the Smith Chart in closer proximity to Γopt (that yields minimum noise). In practice this is done with inductive source degeneration or series feedback. F = Fmin + 3.6 Low-Noise Amplifier Topologies The properties of an LNA are not only determined by the active device and the matching networks around it, but also its topology. A couple of commonplace examples are seen 18 LNA Design Strategy B zS* A zL D Figure 3.10: Smith Chart for matching network design example circuit. 3.7. Determining ΓS and ΓL 19 Figure 3.11: Left: narrowband LNA with inductive source degeneration, right: common source LNA with parallel feedback. in Figure 3.11. Firstly a narrowband LNA with inductive source degeneration and gate inductance for an additional degree of freedom. Secondly a common source LNA with parallel RC feedback for stability purposes. One might also consider a cascade topology with a common base to common source configuration, this improves input-output isolation which simplifies matching. It also improves gain, however the downside with this topology is reduced noise performance. 3.7 Determining ΓS and ΓL When the target specifications are determined, an active device and its working point have been selected, the LNA design [3] then begins with a plot of source stability-, constant noise- and constant gain circles as depicted in Figure 3.12. The required circle equations derivations follow. With the equation for noise in a 2-port amplifier slightly altered with normalized noise resistance and source admittance, it becomes: F = Fmin + rn |ys − yopt |2 . gs (3.14) Expressing ys and yopt in terms of reflection coefficients, with these relations ys = yopt = yields the following expression: 1 − Γs 1 + Γs (3.15) 1 − Γopt 1 + Γopt (3.16) 20 LNA Design Strategy F = Fmin + 4rn |Γs − Γopt |2 . (1 − |Γs |2 )|1 + Γopt |2 (3.17) From this equation it can be seen that the noise figure depends on the variable Γs and three quantities known as the noise parameters: Fmin , rn and Γopt . These quantities are given by the transistor manufacturer but can also be determined experimentally. For a specified noise figure Fi the equation becomes: 2 2 2 Γs − Γopt = Ni + Ni (1 − |Γopt | ) 1 + Ni (1 + Ni )2 (3.18) The centre of the circle is at CF i = Γopt 1 + Ni (3.19) and its radius is rF i = 1 1 + Ni q Ni2 + Ni (1 − |Γopt |2 ). (3.20) The derivations for the constant available gain circles are GA = |S21 |2 1 − |Γs |2 = |S21 |2 ga . S22 −∆Γs 2 1 − | 1−S11 Γs | |1 − S11 Γs |2 (3.21) The centre of the circle is at Ca = ga C1∗ , 1 + ga (|S11 |2 − |∆|2 ) (3.22) where ∗ C1 = S11 − ∆S22 . (3.23) p 1 − 2K|S12 S21 |ga + |S12 S21 |2 ga2 . ra = |1 + ga (|S11 |2 − |∆|2 )| (3.24) The radius of the circle is Conclusively, the output- and input stability circles are plotted with the following equations ∗ ∗ (S22 − ∆S11 ) S12 S21 |=| |, |ΓIN | = 1 2 2 |S22 | − |∆| |S22 |2 − |∆|2 (3.25) ∗ ∗ (S11 − ∆S22 ) S12 S21 |=| |, |ΓOU T | = 1. 2 2 |S11 | − |∆| |S11 |2 − |∆|2 (3.26) |ΓL − |ΓL − 3.7. Determining ΓS and ΓL 21 The centres and radii respectively are CL = ∗ ∗ ) (S22 − ∆S11 |S22 |2 − |∆|2 ∗ ∗ ) (S11 − ∆S22 2 |S11 | − |∆|2 S12 S21 rL = |S22 |2 − |∆|2 S12 S21 . rS = |S11 |2 − |∆|2 CS = (3.27) (3.28) (3.29) (3.30) In the example plot of Figure 3.12 it can be seen that the centre of the constant noiseand constant available gain circles that represent optimum Γs for minimum noise and available gain respectively, do not coincide. It is therefore not possible to achieve a simultaneous optimum match. However, it is possible to shift Γs for available gain by means of inductive source degeneration - without degrading the noise performance significantly. This popular narrowband LNA technique effectively reduces the compromising between noise and available gain when selecting and designing Γs presented to the LNA input. Naturally, if input unconditional stability is desired, Γs has to be situated in the stable region of the Smith Chart. When Γs is selected, the proceedings continue by determining the output reflection coefficient (Γout ) of the LNA and then plot load stability circles. If then Γ∗out is in the stable region, ΓL is set to Γ∗out for a complex conjugate matched output. Should that not be the case, transducer gain circles are drawn to find a ΓL in the stable region that leads to reasonably high transducer gain. 22 LNA Design Strategy Figure 3.12: Smith Chart with constant noise circles (solid) constant available gain circles (dashed) and input stability circle (dotted). C HAPTER 4 2-ports and Noise In [6] a method for computer aided linear 2-port noise analysis is presented. Following traditional 2-port noise analysis, the 2-port is equivalently interpreted as a noiseless 2port coupled with two external noise sources. Depending on the orientation and type of the noise sources, there are six representations of equivalent circuits. However, three representations are sufficient for general applications. These representations are admittance, impedance and chain/cascade as shown in Table 4. The method in [6] introduces that the noise in linear circuits be characterized by correlation matrices as opposed to voltages and currents. Specifically, the correlation matrix of a 2-port consists of four elements that are the power spectrum densities of the cross- and auto-correlations of the noise sources. The noise sources no longer need to be fully correlated or uncorrelated but can also be partially correlated as would be the case with transistor 2-ports. Another advantage is that the analysis explicitly provides data of the noise parameters N Fmin , Rn and Yopt . 4.1 4.1.1 Method of Linear 2-port Noise Analysis Decomposition If the noise analysis is applied to a complete circuit, the initial step is to disassemble it into single component 2-ports. This process is performed in a manner so that one component at a time is isolated from the rest of the circuit, while identifying its relative connection. The disassembly procedure is shown in Figure 4.1. It can be seen that in stage A, the rightmost resistor is isolated and is cascade connected to the remaining circuit. In stage B, the bottommost resistor is isolated and is series connected. Lastly, in stage C, the feedback resistor is isolated from the transistor and is connected in parallel. 23 24 2-ports and Noise Noiseless 2-port i1 u1 i2 Admittance Ci1 i∗1 Ci1 i∗2 CY = ∗ ∗ Ci2 i1 Ci2i2 y y Y = 11 12 y21 y22 Noiseless 2-port u2 Impedance Cu1 u∗1 Cu1 u∗2 CZ = Cu2 u∗1 Cu2 u∗2 z z Z = 11 12 z21 z22 u i Noiseless 2-port Cascade Cu1 u∗1 Cu1 i∗2 CA = ∗ ∗ C i2 u1 Ci2 i2 a a A = 11 12 a21 a22 Table 1: Correlation matrices and electrical matrices of Y, Z and A representations. 4.1.2 Determination of Correlation Matrix Essential to the noise analysis is that the correlation matrix of each basic 2-port is known. That can be determined either theoretically or by noise measurements. Examples of 2ports for which the correlation matrix can be determined theoretically are those that only consist of passive elements. The correlation matrices in impedance and admittance representation are CZ = 2kT <{Z} (4.1) CY = 2kT <{Y } (4.2) where k is the Boltzmann constant, T is the absolute temperature, Z and Y are the electrical matrices in impedance and admittance representation respectively. For active device 2-ports it may be necessary to measure the noise parameters and thereafter calculate the correlation matrix in cascade representation: N Fmin −1 ∗ Rn − Rn Yopt 2 (4.3) CA = 2kT N Fmin −1 − Rn Yopt Rn |Yopt |2 2 4.1.3 Interconnection Once the correlation matrix of each basic 2-port is known, they can be successively interconnected. With each interconnection the resulting correlation matrix is calculated with respect to the representation according to the following: CZ = CZ 1 + CZ 2 (4.4) CY = CY 1 + CY 2 (4.5) 4.1. Method of Linear 2-port Noise Analysis 25 A B C Figure 4.1: 2-port amplifier disassembly illustration. CA = A1 CA2 A∗1 + CA1 (4.6) In the event that the 2-ports are in unmatched representations, e.g. when a series 2-port is connected with one that is in parallel, one has to be converted into the other. Naturally and if applicable, it is preferred to convert to that representation of the next 2-port to avoid redundant calculations. Table 4.1.3 shows a the different transformation matrices. 26 2-ports and Noise Y To Z A From Y Z A 1 0 y11 y12 −y11 1 y21 y22 0 1 −y21 0 z11 z12 1 0 1 −z11 0 −z21 z21 z22 0 1 0 a12 1 −a11 1 0 1 a22 0 −a21 0 1 Table 2: Y, Z and A transformation matrices. 4.1.4 Noise Parameter Calculation In conclusion, when the entire original 2-port is rebuilt its noise parameters can be extracted from the equity N Fmin −1 ∗ Rn − R Y n opt 2 CA = 2kT N Fmin −1 . − R Y Rn |Yopt |∗ n opt 2 C HAPTER 5 The Optimizer A classic approach to LNA design using the Smith Chart has been shown. However, as performance requirements get higher it becomes increasingly difficult to carry out this task manually. Computer aided optimization provides higher efficiency under such circumstances. For that reason a MATLAB implementation of the method described in [6] has been developed at LTU by Johan Borg. It uses stochastical particle swarm optimization (PSO) and has features that enhance the transition to real world applications, including non ideal inductors and real component values. This optimizer has been used to design the EISCAT 3D LNA. 5.1 Description The optimizer takes a user definable LNA topology along with matching networks, where each component is a 2-port with defined properties. Then a corresponding s-parameter and correlation matrix for the entire circuit is computed and passed on to the PSO implementation. Initially in the PSO process, the dimensions of the search space and the solution criteria are defined. Thereafter the population is initialized, that is particle starting positions are either randomized or set by user. Then the swarm is launched to search for an optimum. The velocity and position of the individual particles are updated based on discrete time, current position and velocity, and the particle’s personal best position as well as that of the entire swarm, the global best. The efficiency of PSO is founded in this inter-particle collaboration. This iterative process proceeds until a position has been found that to the greatest possible extent qualify with the predetermined solution criteria. The criteria of this application requires input stability and that noise figure, return loss and gain match or exceed the defined target specifications seen in Table 1. In addition, a list of values is set that specify the desired transfer function. This sets the centre frequency 27 28 The Optimizer Gain NF IRL Centre BW 18.0 dB 0.55 dB 17.0 dB 225 MHz 30 MHz Table 1: Target requirements. and bandwidth of the LNA. As the optimization halts, the applied component values are given. 5.2 Notable functions The optimizer features a set of functions for inter-representation transformation based on on the transormation matrices in Table 2 in Chapter 4. These are used when two representations of different representations are connected. A_to_s A_to_Y A_to_Z s_to_A s_to_Y s_to_Z Y_to_A Y_to_Z Z_to_A Z_to_Y The A-, Z- and Y-representations include electrical- and correlation matrices whereas the s-representation has s-parameters and noise parameters. To attain compliance, the noise parameters are therefore converted (Equation 4.3) into correlation matrix form when the s to A function is used. As for the functions s to Y and s to Z, a passive 2port is implied and the correlation matrices are computed solely based on temperature according to Equations 4.1 and 4.2. In addition there are three functions to connect 2-ports, based on Equations 4.4 - 4.6: A_connect Y_connect Z_connect To initialize an A-representation matrix of individual components while differentiating between parallel and series orientation, the following functions are used: A_s_capacitor A_s_inductor A_s_resistor A_p_capacitor A_p_inductor A_p_resistor 5.3 Usage As previously stated, the optimizer is based on the noise analysis method of [6]. Usage is therefore similar to that described in Chapter 4, except labour intensive tasks are 5.3. Usage 29 conveniently carried out by machine. The one critical step the optimizer will not do however, is to find an appropriate matching network- and amplifier topology. This subject will be addressed in the next chapter. After the LNA circuit to be optimized is known, it has to be loaded into the target function that is subsesequently passed on to PSO. A general case with the inductive source degeneration LNA, displayed in Figure 3.12, would be loaded in the following manner; A_tran=s_to_A(s); % define transistor A_Rs=A_s_resistor(Rs,w); % define series resistor Rs A_Lg=A_s_inductor(Lg,w,IM1); % define series inductor A_Ls=A_p_inductor(Ls,w,IM1); % define parallel inductor A=A_connect(A_Rs,A_Lg); % connect Rs and Lg A=A_connect(A,A_tran); % connect transistor Y=Y_connect(A_to_Y(A),A_to_Y(A_Ls)); % connect Ls A=Y_to_A(Y); % convert to A representation where s is a merged matrix of s- and noise parameters derived from measured data and manufacturer data, w contains frequency domain data and IM 1 is one of four inductor component libraries available. C HAPTER 6 LNA Topology Evaluation Given the complexity of LNA design, it is often times challenging to find a satisfying design topology. It may also have to comply with strict requirements which further increase the difficulty level of this task. In this chapter the single transistor LNA seen in Figure 6.2, that was built for the EISCAT 3D project, is evaluated using the optimizer described in Chapter 5. Then, a simplified version, seen in Figure 6.3 is optimized. After that a version without feedback is evaluated, seen in Figure 6.4. Finally a version with inductive source degeneration, Figure 6.5, is optimized. All variations of the final design have been optimized using the same solution criteria. Furthermore, all circuits have a bandpass filter, Figure 6.1, connected to the output to achieve the required narrowband operation. Lf1 Lf3 Cf1 Cf3 Lf2 Lf4 Cf2 Cf4 Figure 6.1: Bandpass filter. 31 32 LNA Topology Evaluation Cc Cplr Cs4 Cs2 Cl5 Lp1 Ls3 Rd1 Cs1 Rc Cl3 Cl4 Rd2 Ls2 Ls1 Figure 6.2: Circuit 1: Original. Rc Ls2 Cf 2 Rd2 87.97 330.0 22.40 469.5 Cc Lf 1 Cf 3 Lp1 0.1410 68.00 25.19 150.0 Cl4 Lf 2 Cf 4 Cs4 14.06 1.000 373.6 22.00 Cs1 Lf 3 Ls3 Ls 246.1 Cs2 33.00 Lf 4 33.00 Cl3 - 144.0 Ls1 10.00 Cf 1 1.773 Rd1 55.03 15.48 37.13 Table 1: Component values for circuit 1. Capacitances in pF, inductances in nH and resistances in Ω. 6.1 Circuit 1: Original The original EISCAT 3D design, includes the following components: a protective shunt inductor: Lp1 , a directional coupler: Cplr, a protective diode modelled as a capacitor: Cl5 , signal coupling capacitors: Cs1 , Cs4 and Cs2 , bias network decoupling inductors: Ls1 and Ls2 , an input matching network: Ls3 and Cl3 , parallel negative feedback: Rc and Cc , a stabilizing shunt capacitor: Cl4 , series and shunt stabilizing resistors: Rd1 and Rd2 . This circuit performed as expected and the performance parameters are within specifications. Figures 6.6 and 6.7 illustrate the frequency domain behaviour and the optimized component values are shown in Table 1. 6.2 Circuit 2: Simplified In the simplified circuit, a significant number of components are removed from the original circuit. With the exception of those with protective purposes, three signal coupling capacitors and the bias network decoupling inductors. In Figures 6.6 and 6.7, it can be seen that this circuit is unstable for some frequencies. Furthermore, as it can be seen in Table 5, neither NF nor IRL is satisfactory. 6.3. Circuit 3: Without Feedback 33 Figure 6.3: Circuit 2: Simplified. Rc Ls2 Cf 2 Rd2 470.0 33.08 - Cc Lf 1 Cf 3 Lp1 33.00 20.45 68.00 Cl4 Lf 2 Cf 4 Cs4 33.00 952.3 22.00 Cs1 Lf 3 Ls3 Ls 283.4 Cs2 12.00 Lf 4 Cl3 - 353.5 Ls1 12.00 Cf 1 Rd1 339.4 5.243 - Table 2: Component values for circuit 2. Capacitances in pF, inductances in nH and resistances in Ω. 6.3 Circuit 3: Without Feedback Circuit 3 is a copy of the original circuit, short of the parallel feedback. The circuit shows similar in-band characteristics, see Figures 6.6 and 6.7, as that of the original circuit, has equivalent noise- and return loss figures according to Table 5, and roughly 1 dB higher gain. Rc Ls2 Cf 2 Rd2 220.0 15.21 693.5 Cc Lf 1 Cf 3 Lp1 68.00 42.87 150.0 Cl4 Lf 2 Cf 4 Cs4 10.82 2.200 254.7 22.00 Cs1 Lf 3 Ls3 Ls 260.2 Cs2 33.00 Lf 4 33.00 Cl3 - 41.39 Ls1 10.00 Cf 1 1.836 Rd1 43.89 41.93 45.06 Table 3: Component values for circuit 3. Capacitances in pF, inductances in nH and resistances in Ω. 34 LNA Topology Evaluation Figure 6.4: Circuit 3: Without feedback. Figure 6.5: Circuit 4: Inductive source degeneration. 6.4 Circuit 4: Inductive Source Degeneration This circuit employs inductive source degeneration that should shift Γ∗IN closer to Γ∗OP T and thereby render a better noise-gain match possible. According to [2], this method creates a resistive input impedance without the noise of a real resistor. The resistance is controlled by choice of inductance. Since the impedance is only purely resistive at resonance, this method offers a narrowband impedance match. As it can be seen in Figure 6.5, there is no parallel feedback in this circuit. Figures 6.6 and 6.7 indicate that the circuit is unconditionally stable and has a flat in-band gain, much like circuits 1 and 3. In addition, in comparison with the original circuit, this circuit has a notably higher gain and lower noise figure. 6.4. Circuit 4: Inductive Source Degeneration (dB) Circuit 1: Original (dB) 20 0 10 -50 0 -100 -10 a 10GHz -150 35 a 10GHz a 10GHz a 10GHz Circuit 2: Simplified (dB) (dB) 20 0 10 -50 0 -100 -10 a 10GHz -150 Circuit 3: Without feedback (dB) (dB) 20 0 10 -50 0 -100 -10 a 10GHz -150 Circuit 4: Inductive source degeneration (dB) (dB) 20 0 10 -50 0 -100 -10 a 10GHz -150 a 10GHz Figure 6.6: Input return loss (left) and gain (right) of all circuits. A denotes the centre frequency of 225 MHz. 36 LNA Topology Evaluation Circuit 1: Original (dB) (dB) 20 0 10 -50 0 -10 bc 1GHz -100 bc 1GHz bc 1GHz bc 1GHz Circuit 2: Simplified (dB) (dB) 20 0 10 -50 0 -10 bc 1GHz -100 Circuit 3: Without feedback (dB) (dB) 20 0 10 -50 0 -10 bc 1GHz -100 Circuit 4: Inductive source degeneration (dB) (dB) 20 0 10 -50 0 -10 bc 1GHz -100 bc 1GHz Figure 6.7: Input return loss (left) and gain (right) of all circuits. B and C denote the passband 210-240 MHz. 6.4. Circuit 4: Inductive Source Degeneration Rc Ls2 Cf 2 Rd2 220.0 19.33 517.8 Cc Lf 1 Cf 3 Lp1 68.00 38.79 220.0 Cl4 Lf 2 Cf 4 Cs4 12.75 2.200 280.6 22.00 Cs1 Lf 3 Ls3 Ls 160.0 Cs2 22.00 Lf 4 22.00 Cl3 1.646 37 18.70 Ls1 12.00 Cf 1 0.1002 Rd1 68.00 30.94 22.53 Table 4: Component values for circuit 4. Capacitances in pF, inductances in nH and resistances in Ω. Gain NF IRL Target 1 18.0 18.16 0.55 0.5480 17.0 16.95 2 18.47 0.6436 4.796 3 19.33 0.5411 16.96 4 20.07 0.4609 17.04 Table 5: Gain, Noise Figure and IRL of all circuits. C HAPTER 7 Conclusion To design a low-noise amplifier using traditional methods is an involved process that requires a solid understanding of underlying principles. As one performance parameter is affected negatively by the optimization of another it is clear that some degree of compromising is necessary. With high overall performance requirements, little freedom for compromising is allowed and the design process becomes increasingly complex. The computer aid given by the particle swarm MATLAB optimizer has therefore been an invaluable asset. Given an LNA topology and a fitness function to evaluate possible solutions, it outputs component values and performance parameters in terms of return loss, gain and noise figure. The frequency dependance of the former two are also given in graphical format. Since the objective was to find an LNA topology with performance which exceeded that of the EISCAT 3D design, it was natural to use that as a starting point and with the use of the optimizer see how slight changes to it affected performance. An initial run on the basis of the the original circuit showed that it had adequate performace with respect to the predetermined specifications. A simplified and over-constrained toplogy then showed insufficient performance in terms of noise figure and stability although gain was reasonable. In a third run, the effect of the stabilizing feedback stage was examined by removing it from the circuit. As expected this circuit showed higher gain, but could also meet the stability requirements. The conclusion must be that with the existing criteria, feedback is redundant. The final circuit employed inductive source degeneration, where an inductor is connected to the source terminal. The desired effect of this is to decrease the performance loss associated with gain-noise figure compromising. This led to an approximate 2 dB gain increase and 0.1 dB noise figure reduction. The computed value of the source inductor (Ls ) was 1.6 nH, which is low enough to be practically realizable with a micro strip solution. 39 Bibliography [1] EISCAT, “EISCAT 3D the Next Generation European Incoherent Scatter Radar System.” https://e7.eiscat.se/groups/EISCAT_3D\_info/Introduction%20and% 20Brief%20Background.pdf, Nov 2008. [2] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits. Cambridge: Cambridge University Press, 2nd ed., 2004. [3] G. Gonzalez, Microwave Transistor Amplifiers: Analysis and Design. New Jersey: Prentice Hall, 2nd ed., 1996. [4] Avago, “ATF-541M4 Low Noise Enhancement Mode Pseudomorphic HEMT in a Miniature Leadless Package: Data Sheet.” http://www.avagotech.com/docs/ AV02-0924EN, Nov 2008. [5] Maxim, “Impedance Matching and the Smith Chart: The Fundamentals.” http: //www.maxim-ic.com/appnotes.cfm/an_pk/742, Nov 2008. [6] H. Hillbrand and P. H. Russer, “An efficient method for computer aided noise analysis of linear amplifier networks,” IEEE Transactions on Circuits and Systems, vol. CAS23, no. 4, pp. 235–238, 1976. 41