Small-Signal and Low-Noise Amplifier Design

Introduction and Context

RF receivers and transmitters routinely process signals whose amplitudes are only a few microvolts to a few millivolts. An RF amplifier must therefore provide gain without destroying information through distortion, instability, or excessive noise. At postgraduate level, amplifier design is best understood as a simultaneous optimization of gain, noise, linearity, stability, bandwidth, power consumption, and manufacturability.

A small-signal amplifier operates around a bias point, or quiescent point, where the variations in voltage and current caused by the input signal are sufficiently small that the device can be represented by a linearized model. A low-noise amplifier (LNA) is the first active stage of a receiver and is optimized primarily to achieve a low noise figure while providing adequate gain, input matching, reverse isolation, linearity, and stability.

The design problem is inherently multi-objective. The source impedance that gives minimum noise is generally different from the impedance that gives maximum available power gain. Consequently, practical LNA design requires controlled compromise rather than optimization of one parameter in isolation.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Linearize RF transistor models about a bias point and interpret their small-signal parameters.
2. Calculate gain, input and output impedances, and frequency response using hybrid-π or S-parameter models.
3. Explain noise sources, noise figure, equivalent noise temperature, and noise circles.
4. Apply the Friis cascade formula to multistage receiver design.
5. Evaluate unconditional and conditional stability using S-parameters.
6. Design a practical narrowband LNA subject to gain, noise, matching, stability, and linearity constraints.
7. Recognize the limitations of unilateral approximations and simulation-only design.

1. Small-Signal Representation of RF Transistors

1.1 Bias point and linearization

For a nonlinear device, the drain current may be written as a function of terminal voltages. For a field-effect transistor,

At the operating point, small changes can be approximated by a first-order Taylor expansion:

where g_m = ∂I_D/∂V_GS is the transconductance in siemens and g_ds = ∂I_D/∂V_DS is the output conductance in siemens. The approximation is valid when the signal excursion is small enough that second-order and higher-order terms are negligible.

For a bipolar junction transistor, the corresponding controlled-source relation is approximately ∆i_c = g_m∆v_be, with g_m = I_C/V_T. Here I_C is collector current and V_T is thermal voltage, approximately 25.9 mV at 300 K.

1.2 Hybrid-π model

The hybrid-π model represents a transistor by a controlled current source, input resistance or conductance, output conductance, and parasitic capacitances. In a common-source MOSFET model, important elements include:

The high-frequency behavior is dominated by these capacitances and by package, interconnect, and substrate parasitics. The feedback capacitance C_gd is particularly important because it couples output variations back to the input and produces the Miller effect in voltage amplifiers.

For a common-source stage with a resistive drain load, neglecting feedback and finite output resistance, the midband voltage gain is approximately

where R_D is the effective drain load. If output resistance is significant, the expression becomes

The parallel operator means R_D ∥ r_o = R_Dr_o/(R_D+r_o). The formula assumes a low source impedance, negligible source degeneration, and operation in the midband where capacitive reactances do not dominate.

1.3 Source degeneration

An unbypassed source resistance improves linearity and stability but reduces gain. In a simplified low-frequency model,

where R_S is the source-degeneration resistance. At RF, an inductive source degeneration element can simultaneously contribute gain stabilization, input matching, and noise optimization. This is one reason inductively degenerated CMOS LNAs are widely used in narrowband receivers.

2. RF Gain and Matching

2.1 Power gain definitions

RF design distinguishes several gains:

• Available gain: power available from the two-port divided by power accepted from the source.
• Operating power gain: power delivered to the load divided by power accepted at the input.
• Transducer gain: power delivered to the load divided by power available from the source.
• Unilateral gain: gain calculated by assuming reverse transmission is zero.

For a two-port described by S-parameters, with source reflection coefficient Γ_S and load reflection coefficient Γ_L, the transducer gain is

This expression assumes reference impedances are known, usually 50 Ω, and that the network is linear at the signal level under consideration. Reflection coefficients must satisfy |Γ| ≤ 1 for passive source and load terminations.

If reverse transmission is negligible, S₁₂≈0, the expression simplifies to the unilateral form

The unilateral approximation is useful for preliminary design but can be seriously inaccurate when S₁₂ is appreciable.

2.2 Conjugate matching

For a passive source with impedance Z_S, maximum power transfer occurs when the amplifier input impedance is its complex conjugate:

Likewise, maximum power delivery to a load occurs when the output impedance is conjugately matched to the load. In an LNA, however, the impedance for minimum noise is usually not identical to the impedance for maximum gain. Matching networks are therefore designed using a compromise selected from gain circles, noise circles, stability circles, and system requirements.

2.3 Narrowband matching networks

At a single frequency, lumped inductors and capacitors can transform a transistor's complex impedance to 50 Ω. A series inductor can cancel capacitive reactance because

where f is frequency in hertz, L inductance in henries, and C capacitance in farads. Resonance occurs when X_L + X_C = 0, giving

This ideal relation excludes loss and parasitic elements. At microwave frequencies, layout inductance, pad capacitance, transmission-line sections, and electromagnetic coupling must be included in the effective values.

3. Noise in RF Amplifiers

3.1 Physical noise mechanisms

Important noise mechanisms include:

• Thermal noise in resistors and lossy conductors.
• Shot noise due to discrete charge flow in junction devices.
• Flicker noise, or 1/f noise, caused by trapping and release phenomena.
• Induced gate noise in high-frequency MOSFETs.
• Partition and avalanche noise in some bipolar and high-field devices.
• Noise from bias networks, substrate coupling, and passive components.

The mean-square thermal noise voltage of a resistor over bandwidth B is

where k is Boltzmann's constant in J/K, T is absolute temperature in kelvin, R is resistance in ohms, and B is bandwidth in hertz. The corresponding available noise power from a matched resistor is kTB watts. The equation assumes thermal equilibrium and a frequency range over which the resistor is adequately modeled as frequency independent.

3.2 Noise factor and noise figure

Noise factor is defined as

where signal-to-noise ratios are measured under specified source and load conditions. Noise factor is dimensionless; noise figure is its decibel representation:

An ideal noiseless amplifier has F = 1 and NF = 0 dB. Real amplifiers have F > 1.

The equivalent input noise temperature is related to noise factor by

where T₀ is the reference temperature, conventionally 290 K. This representation is especially useful for satellite and radio-astronomy systems.

3.3 Noise parameters

For a two-port at a specified frequency, noise behavior is commonly described by four parameters: minimum noise factor F_min, optimum source reflection coefficient Γ_opt, equivalent noise resistance R_n, and reference impedance Z₀. A common expression is

The exact form depends on the adopted normalization and whether R_n is expressed in ohms or normalized units. The equation shows that noise factor increases as the source reflection coefficient moves away from the optimum value. Noise circles are contours of constant F on a Smith chart.

Noise matching should not be confused with power matching. A design that forces Γ_S = Γ_opt may sacrifice gain or bandwidth; a practical design selects an acceptable point on the gain–noise–stability trade-off surface.

3.4 Cascaded amplifiers: Friis formula

For a cascade of stages, the total noise factor is

where F_i is the linear noise factor of stage i and G_i is the available power gain of stage i, both dimensionless. The formula assumes matched interconnections and stable, linear stages. It demonstrates why the first receiver stage has dominant influence: gain before a noisy later stage suppresses that later stage's contribution.

4. Stability and Oscillation

An amplifier can oscillate when feedback and impedance interactions cause the loop to satisfy both magnitude and phase conditions for oscillation. Stability must therefore be checked for all passive source and load impedances, not only for a nominal 50 Ω test environment.

For a two-port, define

and the Rollett stability factor

For nonzero S₁₂S₂₁, unconditional stability at a frequency requires

These conditions are based on linear S-parameters and passive terminations. They do not guarantee large-signal stability, stability under bias transients, or stability across frequencies not measured or simulated.

If the conditions are not met, the device is conditionally stable. Stabilization methods include resistive feedback, source degeneration, lossy matching networks, gate or base series resistance, and careful bias decoupling. Each method usually degrades some combination of gain, noise figure, or efficiency.

5. Linearity and Dynamic Range

Noise performance alone is insufficient. Strong interferers can compress the desired signal or generate intermodulation products. The 1-dB compression point is the input or output power at which gain falls 1 dB below its small-signal extrapolation.

For a two-tone test with input frequencies f₁ and f₂, third-order products occur at 2f₁-f₂ and 2f₂-f₁. The input third-order intercept point, IIP3, is an extrapolated metric, not normally a directly reached operating point. For a simplified equal-tone model, if fundamental output power is P_F and third-order product power is P_IM3, then

when all powers are expressed in dBm and P_in is the input tone power in dBm. The relation assumes the fundamental and third-order responses follow slopes of 1 and 3, respectively, over the extrapolation range.

Receiver dynamic range is limited at the lower end by noise and at the upper end by compression and intermodulation. A low-noise design with insufficient linearity may perform poorly in a dense spectral environment.

6. Practical LNA Design Procedure

A disciplined workflow is more reliable than tuning individual components without system context.

Step 1: Define specifications

Specify frequency or band, supply voltage, gain, maximum noise figure, input and output return loss, bandwidth, IIP3 or compression point, current consumption, and allowable unconditional-stability margin.

Step 2: Select device and bias point

Use measured or vendor-provided S-parameters and noise parameters at the intended bias. Bias affects transconductance, capacitances, noise, linearity, and breakdown margin. Data measured at a different bias cannot safely be substituted without verification.

Step 3: Establish stability

Calculate K and |∆| across frequency. Include bias networks, package models, bond wires, and probable source and load impedances. Add stabilization only where necessary because indiscriminate resistance increases noise and reduces gain.

Step 4: Choose the source match

Plot noise circles and available-gain circles. Select Γ_S close enough to Γ_opt to meet noise requirements while preserving gain and stability. The source matching network is then synthesized and transformed into realizable component values.

Step 5: Choose the load match

Select Γ_L for desired gain, output return loss, bandwidth, and linearity. A narrowband conjugate match can provide high gain but may have poor tolerance and limited bandwidth.

Step 6: Simulate with progressively realistic models

Begin with ideal lumped components, then add transistor parasitics, package models, layout-extracted parasitics, substrate losses, and electromagnetic coupling. Perform nominal, corner, Monte Carlo, and temperature simulations.

Step 7: Layout and verify

Keep RF paths short, use a continuous ground reference, place bypass capacitors close to bias pins, isolate input and output routing, and provide multiple low-inductance ground vias. After fabrication, de-embed fixtures where necessary and compare measured data with the model.

7. Worked Design Example: Narrowband CMOS LNA

Consider an illustrative 2.4 GHz receiver front end using a 50 Ω system. The required small-signal gain is at least 15 dB, noise figure below 2.5 dB, and supply current below 12 mA. At the selected bias, assume measured data indicate S₂₁=4.5∠(-20°), S₁₁=0.62∠(-80°), and a minimum noise factor of 1.45 at the optimum source reflection coefficient. Assume a first-stage gain of 15 dB and a second-stage gain of 10 dB with noise figures of 2.2 dB and 5 dB, respectively.

First convert decibel quantities to linear factors. For power gain,

For the noise factors,

Using Friis for the two stages,

Therefore,

The cascade meets the 2.5 dB target. If the first stage gain were only 5 dB, its linear gain would be 3.162 and the total factor would become approximately 2.344, or 3.70 dB. This illustrates why the first LNA must provide useful gain, not merely minimum noise.

At 2.4 GHz, suppose the input capacitance to be resonated is 0.8 pF. The ideal inductance that resonates with it is

The value is only a starting point. Pad capacitance, transistor feedback, inductor self-resonance, quality factor, and source degeneration alter the final value. Electromagnetic simulation and measurement are required before production release.

8. Applied Case: Receiver LNA in a Dense Wireless Environment

Background. An illustrative wireless sensor receiver operates at 2.4 GHz near high-power Wi-Fi transmitters. The initial prototype achieves 20 dB gain and 1.8 dB noise figure in a quiet laboratory.

Problem. In a populated test environment, sensitivity is worse than predicted and the receiver exhibits desensitization. Two-tone testing reveals third-order intermodulation products inside the channel.

Analysis. The LNA was optimized for noise and nominal gain, but its source degeneration was too small and the bias network allowed supply coupling. The input match also shifted when the antenna impedance changed. Although the nominal S-parameter simulation showed K > 1 at the center frequency, stability margin degraded near the band edges with realistic loads.

Approach. The design team introduced moderate inductive source degeneration, improved supply filtering, separated input and output ground-return paths, and selected a slightly less aggressive noise match. A small amount of controlled feedback improved reverse isolation and broadband stability. The matching network was redesigned using measured noise parameters and antenna impedance statistics rather than a fixed 50 Ω assumption.

Outcome. The revised design sacrificed approximately 0.2 dB of nominal noise figure but improved IIP3, load tolerance, and out-of-band stability. System sensitivity improved in the actual interference environment because the receiver no longer suffered severe desensitization.

Learning points. The best laboratory noise figure is not necessarily the best system design. Environmental interference, antenna mismatch, stability margin, and linearity must be evaluated together. Receiver specifications should be derived from field conditions, not only from ideal bench measurements.

9. Limitations, Reliability, and Professional Practice

Small-signal analysis is local. It does not predict compression, breakdown, self-heating, memory effects, or large-signal oscillation. S-parameters are also dependent on bias, temperature, calibration plane, and measurement accuracy. Noise parameters may vary considerably with frequency and bias, and vendor data may omit uncertainty information.

Designers should distinguish simulated precision from physical accuracy. A result reported to many decimal places is not necessarily reliable if component tolerances, model error, and layout parasitics dominate. Calibration, uncertainty analysis, and correlation with prototypes are essential professional practices.

RF design also has regulatory implications. Oscillation or unintended radiation can violate electromagnetic compatibility requirements. Excessive transmit leakage can interfere with safety-critical systems. Designers must comply with applicable spectrum regulations, electromagnetic compatibility standards, semiconductor reliability limits, and product safety requirements.

10. Contemporary Developments

Modern LNAs increasingly use advanced CMOS, SiGe BiCMOS, GaN, and compound-semiconductor technologies according to frequency, noise, power, and linearity requirements. Digitally assisted calibration can compensate for process and temperature variation, while tunable matching networks support multiband operation. Machine-learning methods are being explored for surrogate modeling, layout optimization, and yield prediction; however, such methods do not replace physical constraints, measurement, or stability analysis.

Integrated receivers also increasingly use differential architectures, transformer coupling, current reuse, noise cancellation, and frequency-selective feedback. These techniques can improve noise or power efficiency but introduce additional common-mode, coupling, and robustness concerns. The fundamental design principles remain unchanged: define the system objective, use valid models, evaluate trade-offs, and verify performance under realistic conditions.

Best-Practice Checklist

• Use linear units for calculations involving gain and noise factor; convert to dB only for reporting.
• Never assume noise matching and power matching are identical.
• Check stability across frequency and for realistic passive source and load impedances.
• Include package, pad, interconnect, and bias-network parasitics.
• Evaluate gain, NF, IIP3, compression, return loss, current, and tolerance together.
• Validate simulations with calibrated measurements and appropriate de-embedding.
• Maintain adequate layout isolation between input and output.
• Treat AI-based optimization as a design aid, not as a substitute for engineering judgment.

Chapter Summary

Small-signal RF amplifier design begins by linearizing a transistor about its bias point. The resulting model provides transconductance, output resistance, capacitances, and feedback paths needed to estimate gain and impedance. At microwave frequencies, S-parameters are usually more practical than low-frequency circuit parameters for describing a complete two-port.

An LNA must minimize noise while maintaining useful gain, linearity, bandwidth, stability, and acceptable power consumption. Noise factor and noise figure quantify degradation in signal-to-noise ratio, while noise parameters describe how source impedance affects noise. The Friis formula explains why gain and low noise in the first receiver stage are especially important.

Matching networks transform source and load impedances, but the optimum noise match generally differs from the maximum-gain match. Stability must be evaluated using S-parameters and must include realistic frequency and termination conditions. Finally, successful designs require iterative simulation, layout-aware modeling, fabrication, calibrated measurement, and system-level evaluation in the intended interference environment.

Key Terms

Small-signal model: Linearized representation of a device around its bias point.

Bias point: Quiescent voltage and current operating condition.

Transconductance: Change in controlled current per unit change in controlling voltage.

Output conductance: Change in output current per unit output-voltage change.

Hybrid-π model: High-frequency transistor equivalent circuit using controlled sources and parasitic elements.

S-parameters: Traveling-wave network parameters used to characterize RF two-ports.

Transducer gain: Power delivered to the load divided by power available from the source.

Noise factor: Ratio of input to output signal-to-noise ratios.

Noise figure: Noise factor expressed in decibels.

Equivalent noise temperature: Temperature representation of added noise.

Noise matching: Selection of source impedance for a desired noise factor.

Conjugate match: Condition in which load impedance equals the complex conjugate of source impedance.

Rollett stability factor: S-parameter metric used to assess unconditional stability.

Reverse isolation: Suppression of signal transmission from output to input.

1-dB compression point: Power level at which gain is 1 dB below its linear extrapolation.

IIP3: Input third-order intercept point.

Friis formula: Expression for noise factor of cascaded stages.

Source degeneration: Feedback introduced through an impedance in the source or emitter path.

De-embedding: Removal of fixture and interconnect effects from measured data.

Unconditional stability: Stability for all passive source and load impedances.

Review Questions

Short Answer Questions

1. What assumptions justify the use of a small-signal transistor model?
2. Define noise factor and noise figure.
3. Why is the first stage of a receiver especially important in the Friis formula?
4. What is the physical significance of C_gd in a common-source amplifier?
5. State the two conditions commonly used for unconditional stability using the Rollett factor.

Long Answer Questions

1. Explain the hybrid-π model of an RF transistor and derive the approximate midband gain of a common-source stage.
2. Discuss the difference between power matching and noise matching, including the role of noise circles.
3. Explain the major noise mechanisms in RF amplifiers and compare noise factor with equivalent noise temperature.
4. Describe a complete narrowband LNA design workflow from specification to measurement.
5. Discuss how gain, noise figure, linearity, stability, bandwidth, and power consumption interact in practical LNA design.

Analytical / Application Questions

1. A first amplifier has a gain of 12 dB and noise figure of 2 dB. A second amplifier has noise figure of 6 dB. Calculate the contribution of the second stage to total noise factor and explain how increasing first-stage gain changes it.
2. An LNA has S₁₁=0.5, S₂₂=0.4, S₁₂=0.05, S₂₁=3, and real-valued parameters for this approximation. Calculate ∆ and K, then assess whether the simple unconditional-stability test is satisfied.
3. Design a conceptual matching strategy for an LNA whose optimum noise reflection coefficient differs substantially from its maximum-gain reflection coefficient. State which plots, measurements, and trade-offs should guide the final choice.

Multiple Choice Questions with Answers

Question 1

Which quantity is the ratio of input signal-to-noise ratio to output signal-to-noise ratio?

A. Available gain

B. Noise factor

C. Transducer gain

D. Return loss

Correct Answer: B. Noise factor

Question 2

For two cascaded stages, which stage most strongly affects total noise performance when its gain is substantial?

A. The first stage

B. The last stage

C. Only the stage with the highest supply voltage

D. Only the stage with the lowest bandwidth

Correct Answer: A. The first stage

Question 3

The condition for maximum power transfer to a load is generally

A. ZL = ZS

B. ZL = 1/ZS

C. ZL = ZS*

D. ZL = 0

Correct Answer: C. ZL = ZS*

Question 4

What does the parameter gm represent in a transistor small-signal model?

A. Output resistance

B. Transconductance

C. Gate capacitance

D. Noise temperature

Correct Answer: B. Transconductance

Question 5

If a noise factor is 2, its noise figure is approximately

A. 0.5 dB

B. 2 dB

C. 3.01 dB

D. 6 dB

Correct Answer: C. 3.01 dB

Question 6

Which pair of conditions is commonly used for unconditional stability with the Rollett factor?

A. K < 1 and |∆| > 1

B. K = 0 and |∆| = 0

C. K > 1 and |∆| < 1

D. K > 0 and |∆| > 1

Correct Answer: C. K > 1 and |∆| < 1

Question 7

A resistor’s available thermal-noise power over bandwidth B at temperature T is

A. kTB

B. 4kTRB

C. kT/B

D. 4kT/RB

Correct Answer: A. kTB

Question 8

The 1-dB compression point indicates that

A. Noise figure has increased by exactly 1 dB

B. Gain is 1 dB below its small-signal extrapolation

C. Input return loss is 1 dB

D. The amplifier has unity gain

Correct Answer: B. Gain is 1 dB below its small-signal extrapolation

Question 9

Why is source degeneration often used in an LNA?

A. It can improve linearity and stability

B. It eliminates all transistor capacitances

C. It guarantees zero noise figure

D. It makes reverse transmission infinite

Correct Answer: A. It can improve linearity and stability

Question 10

In the Friis formula, the gain of an earlier stage reduces the contribution of later-stage noise because it

A. Raises the physical temperature of later stages

B. Divides later noise factors’ excess terms

C. Makes all later stages noiseless

D. Removes the source impedance

Correct Answer: B. Divides later noise factors’ excess terms