0
\$\begingroup\$

I am familiar with circuit design - Basic Op-Amp Design, concepts like stablity, Phase margin and I am a beginner in Microwave Engineering. I am familiar with some basic ideas and little familiar with devices like pHEMT (High Electron Mobility Transistor) ,mHEMT,HBT and working principles.

I have a task to give a presentation on this Paper (patent) which can be downloaded here.

One of the circuit diagram: enter image description here

The title of the Patent is Switchable Multiband LNA Design. This patent talks about designing a Low Noise Amplifier. On the whole I understand the circuit, which consists of a cascode transistors to avoid the miller effect and the feedback (Which also gives some stability problems). I can understand the circuit theory part of it but I wanted to really understand the Radio Frequency Engineering used in this patent - but I am stuck with some basic questions like:

  • What are the challenges in designing Amplifiers with high frequencies? - Is it not possible with popular technologies like CMOS - why are they using something like Si/Ge or InGaAs?
  • Above which frequencies does the amplifier design becomes difficult? How far does the normal Op-Amp with Differential stage work? (I usually design it from DC-some MHz)
  • What basic knowledge is required in LNA Design? Why can't we use the same approach that we do in a normal Amplifier design? Like if we directly connect wires, impedance won't match? What is that 50 Ohms matching anyways? Why can't other others other than 50 Ohms can be used for matching? Its just we need matched resistances right?

  • What is noise matching?

  • Why in RF (Radio frequency) -MMIC (monolithic microwave Integrated circuits), few transistors are used and why are they usually big?

I tried to read this paper but concepts were just totally new and wierd to me - like using spiral inductors, biasing for transistors, variable capacitor. ]

If anybody could recommend a text book which could provide a basic knowledge for this given paper, it would be helpful.

\$\endgroup\$
  • 1
    \$\begingroup\$ If my attempts related to this topic are any indication, this question is far too broad and needs to be broken up. \$\endgroup\$ – Reinderien May 23 '18 at 21:42
  • \$\begingroup\$ Hint on the first question: What does the "HEM" in pHEMT stand for? \$\endgroup\$ – The Photon May 23 '18 at 22:00
  • \$\begingroup\$ @ThePhoton Its High Electron Mobility Transistor. \$\endgroup\$ – sundar May 23 '18 at 22:07
  • \$\begingroup\$ @sundar, so does that help answer your first question? \$\endgroup\$ – The Photon May 23 '18 at 22:39
  • \$\begingroup\$ @ThePhoton I can somehow relate it - High frequencies require high mobility and better noise characteristics - may be thats why they use HEMT - which can be realized in Si/Ge technolgies. \$\endgroup\$ – sundar May 24 '18 at 22:20
2
\$\begingroup\$

1) what are the challenges in designing amplifiers for high frequencies?

Lets split that question into 2 parts

------narrowband design

A narrowband design, with low phasenoise Local Oscillator, might have 100Hz bandwidth; NASA communicates with some satellites at such low bandwidths (low data rates); in a 50_ohm system at room temperature the noise floor will be 0.89 nanoVolts RMS * sqrt(bandwidth), and a 100Hz system (which will support approximately 100 bits per second data rate) has noise of 0.89nV * sqrt(100) or 0.89nV * 10 or 8.9 nanoVolts RMS noise floor. With Shannon limit being near 0dB SNR, the recoverable signal energy will also be 8.9 nanoVolts RMS signal. To feed the signal into a quantizer/digitizer/ADC at 1voltRMS (2.828v PP) you will need a voltage gain near 300 Million, without oscillation.

Yet VDD distribution circuitry (providing power for each of perhaps 20 stages of gain or mixing or variable-gain (to provide adjustable gain to maintain near full-scale ADC loading)) will have resonances, and you must prevent the VDD wiring being a sneak feedback path.

You also need to select a precise small range of frequencies; that requires you learn about phasenoise behaviors, because angular noise of Local Oscillators will mix with close_in and far_out "noise" (random and periodic), degrading the SNR; further, to select a small range of frequencies, you need sharp filters either by cavities or striplines or LC or active filters (usually with noise floor far above default 50 ohm floors of passive filters)

------broadband (DC --- light, 1MHz to 100GHZ) design

In broadband design, if you need FLAT frequency response, then any resonances (unless properly dampened) will be a problem; and there are several methods to cause resonances

a) modes of energy storage in the cavities of any shielding structures

b) inductance + capacitance energy storage

c) stripline energy: power strips, signal_in strips, signal_out strips, GND strips (2milliMeters of metal causes 2mm of delay and also stores energy)

d) the VDD distribution "tree"

Also broadband design is very vulnerable to coupling between input and output, yet the tuned circuits are not permitted to reject unwished energy.

2) above what frequency does "normal opamp" no longer work?

There are at least three tradeoffs here

a) a diffpair with emitter-follower (the Cherry-Hopper circuit) may be used at quite high frequencies, because the total phase shift is very small

b) inductances become a big problem, partly because GND becomes an uncontrolled node

c) 10pF capacitance at 1 volt RMS at 10GHz requires what power?

Power = F * C * Vrms*2 = 10^10 Hertz * 10^-11 Farad * 1voltRMS^2 = 0.1 watt

Thus an opamp driving 10pF at 10GHz at 1 volt RMS must provide 100 milliWatts to the load.

3) the ART of LNA design : distortion, Zin, Zout, Noise Floor, phase shifts that upset data eyes, VDD filtering, ESD survival, instability (oscillation) for various Zsource or Zload

4) noise matching: because signal energy becomes so precious (gain becomes so difficult to achieve), we seek conjugate matches between stages;

5) why are only a few large transistors used: we cannot use the standard OPAMP because the UGBW may be only 1,000MHz or 2,000MHz; we accept we cannot use large ratios of feedback resistors to achieve precise gain because parasitic C across the resistors lead to capacitive-division setting the gain, not the resistor ratios; we use a few large transistors, operating at HIGH CURRENTS to provide HIGH TRANSCONDUCTANCE, perhaps with emitter resistors to linearize the stage gain and push IP2 and IP3 up to high intercept levels for satisfactory distortion; for channelized communications such as cellphones, the IP3 behavior is key

\$\endgroup\$
2
\$\begingroup\$

What are the challenges in designing Amplifiers with high frequencies?

This depends on what you consider high. The challenging thing is always shifting. Where it used to impressive to get a few GHz out of standard CMOS, currently people are working CMOS circuits well into the millimeter wave spectrum - Oscillators at 500 GHz or higher, 200 GHz downconverters, etc.

There are two parts to this. First, there is the natural bandwidth limitation of circuits: the capacitance due to interconnections, output load, gate-bulk capacitance, etc. limit the upper frequency we can get in "classic" analog design. The way RF designers get around some of these issues is by inserting inductors to tune out the capacitance. This naturally also forms a band-limited design.

Oscillation also becomes a problem. In lower frequency analog design, the source of the oscillation is our own feedback network. However, in high frequency amplifiers, the capacitance from output to inputs (for example, the drain-gate capacitance) can provide sufficient feedback to result in an unstable amplifier. Usually this is not visible because at those frequencies the output capacitance prevents large gain - but as I just said, now this output capacitance is tuned out and becomes invisible.

Another difference is that layout and simulation becomes more important as we increase in frequency. At lower frequencies and general analog integrated circuit design, a simply parasitic extraction can give accurate results. When we go up in frequency, this is no longer the case, and we require full 3D electromagnetic simulations of things like the network connecting the terminals of a transistor to the surrounding metal layers.

\$\endgroup\$

Not the answer you're looking for? Browse other questions tagged or ask your own question.