Small and large signal S parameters

Am 09.02.19 um 12:36 schrieb snipped-for-privacy@gmail.com:

Yes. It's currently Gospel that large signal s-parameters are required for oscillator design. Oscillators are inherently nonlinear since there must be some limiting / gain control or the amplitude would collapse or explode otherwise.

I don't see that so extreme. I got the the point where the crystal phase noise is all that matters without HB.

In the Agilent world, look for x-parameters.

Here are some papers from Rohde; he is a mass publisher so it will be redundant. There is more interesting material on the synergy microwave web site.

Older s-parameters should be small signal, esp. when there is only bias given and no levels.

I'm just trying to determine myself if I could use harmonic balance to simulate the noise behavior of a chopper amplifier or if that is too non-linear. I'm still struggling with importing spice models.

cheers, Gerhard

The first thing an oscillator has to do is to start up. It's initially in the small signal condition, so the resonator + small signal S params have to be unstable. So you can't ignore them.

Gerhard knows more RF than I do, so I'll let him carry on. ;)

Cheers

Phil Hobbs

I'd just like to add that small-signal analysis can be very useful in large-signal situations. For example, I'm working on designing high-voltage amplifiers with power MOSFETs. These are capable of fast slewing and high power levels, with excursions of hundreds of volts, and high currents into capacitive loads. At first it appeared that small-signal measurements and analysis would not be so useful. But then I realized that when the circuit was non-linear, slewing, delivering current, I'd use the appropriate analysis, i = C dV/dt, etc., but after it was done slewing, it'd be the steady-state condition that was most important, and most of my effort went into solving the equations for that, and using them to make the design rock solid. I made sure the amplifier would then be operating class-A. This scheme worked well, and I created a simple inexpensive design for a 1200-volt DC power amplifier that has a -3dB bandwidth of 1 MHz. It's gratifying to see the amplifier perform well, fast, powerful, yet stable, with low measured phase shifts at 1MHz and beyond, so that with a transducer, it can work well in a wideband servo. Small-signal analysis rocks!

BTW, at these frequencies, I wasn't using s-parameters.

Phil Hobbs wrote in news: snipped-for-privacy@supernews.com:

In switch mode power supply design and use, this is known as "slow start" of "hard start". If one punches the start of some oscillators too hard the initial swing does not start and the oscillator 'latches up' and the circuit start fails. In many cases that input needs to be dampened or led or lagged to achieve a "soft start" condition that ensures that the oscillator always starts.

Think about the Barkhausen criterion, the total loop gain must be larger than 1 and the total phase around the loop must be correct.

At oscillator startup, there is always some amplifier input related white thermal noise. The amplifier amplifies the noise G times. This noise is them coupled back to the input via the frequency selective network with quality Q with correct phase, thus a narrow band noise with bandwidth f/Q is connected back to amplifier input. The noise is

This continues, until the signal amplitude is limited by the available voltage swing.

The small signal s-parameters are initially critical, since the Barkhausen criterions must be satisfied from thermal noise levels up to limiting.

If the amp's transfer function has the right (ie, wrong) shape you can make an oscillator that will run but not start.

Yup. The math isn't hard, and (when both approaches apply) one formula has more information than a week's worth of simulations.

Cheers

Phil Hobbs

Biasing the amplifier into class-C and the oscillator will not automatically start. Once started the amplification in class C will kick the resonator once each cycle.

In practice, such oscillators may start by quickly applying the operating voltage, which allows some collector current to flow momentarily. You may end up with an oscillator, which starts nicely when battery powered but not when mains powered through linear power supply with big electrolytes :-). With such power supply, the voltage may start too slowly to initiate the initial cycle.

It certainly is the Gospel to use large signal S parameters, but things are changing rapidly. A new breed of bi-junction transistors(e.g., HFA3134 - Renesas Semiconductor) with a transition frequency of 8 GHz(well in the RF/microwave range) DOES NOY require the use of S parameters at all. The easily downloadable data sheet does not contain any S parameters. I have used it to SPICE simulate common emitter/base feedback/negative resistance oscillators upto 2 GHz. No issues at all.

My design scheme is simple.

1. AS a real-world oscillator cannot be expedted to dump all the signal energy at the design frequency, define a set of tolerances (e.g., 5%) on the first n (3 - 4) harmonics.
2. Compute passive component values e.g., for a common emiiter feedback oscillator, with a simple C program (to remove silly calculation errors) and then SPICE simulate it with transientr analysis.
3. Fourier transform transient analysis output, and compute power spectrun and check if the frequencies corresponding to the first n(3 - 4) highest peaks fall within the predefined tolerances. If not, adjust resonator component values and iterate through the above steps till tolerances ae satisfied - convergence. If yes, task is complete.

I do have a copy of Rohde's 2005 book on microwave oscillator design, but suffers from the same problem as books on this topic by others(Grebennikov, Ludwig Bretchko, Pozar etc.,) hundreds of pages of theory. We do have ADS at work, but most of us think that the learning curve is very steep, and the results are very non-intuitive.

Thank you very much for the URLs _ I will definitely look through them.

It depends on the type of oscillator one is looking at. A differential oscillator, by its very nature, does not need any transistor biasing, and so it does not matter if the S parameters are used or not. In addition, for the new breed of transistors(e/g/. HFA3134 from Renessas Semiconductor) the datasheet DOES NOT list ANY S parameters. With a fT of 8 GHz, it would work very well in the RF - microwave frequency range.

Sure. It has a super-detailed SPICE model for both the device and the package parasitics, which is going to be more useful than S parameters. You can generate S parameters from the model, but going the other way is a lot more complicated.

Cheers

Phil Hobbs

Instant-start oscillators are fun.

The challenge is to make every cycle, including the first one, exactly periodic.

Anything really interesting is going to be nonlinear, so may as well Spice.

Am 11.02.19 um 19:06 schrieb John Larkin:

Unfortunately, nonlinear analysis in Spice is quite meagre. You just get transient analysis and that's it. OK, add FFT from the post processor.

For noise and frequency response analysis, the circuit is linearized around the operating point, so it is small signal only by definition.

No nonlinear noise, no harmonic balance, no large signal frequency response.

I would not get very far determining the noise characteristics of my chopper amplifiers. How could it linearize the circuit in the presence of the chopp clock and the continuous switching?

How do I determine the noise level of an amplifier that is near compression? That is the normal case in the sustaining amplifier of an oscillator.

How could I see the jitter induced by noise or self heating? Transient simulation is noise free.

How do I simulate a SRD frequency multiplier? Cannot. There is no concept of carrier lifetime in Spice. No PIN diodes. Oh, 1N4007 is a PIN diode. (the others in the series aren't)

cheers, Gerhard

I have added time-domain noise sources to LT Spice, but that's hack. Spice is great for large-signal, time-domain stuff.

You had me scared for a minute there. A stock 1N914 sure behaves like an SRD. Probably unrealistically so.

Version 4 SHEET 1 880 680 WIRE 112 128 80 128 WIRE 176 128 112 128 WIRE 304 128 240 128 WIRE 320 128 304 128 WIRE 80 160 80 128 WIRE 320 176 320 128 WIRE 80 288 80 240 WIRE 320 288 320 256 FLAG 80 288 0 FLAG 320 288 0 FLAG 112 128 IN FLAG 304 128 OUT SYMBOL diode 176 144 R270 WINDOW 0 -42 34 VTop 2 WINDOW 3 -53 33 VBottom 2 SYMATTR InstName D1 SYMATTR Value 1N914 SYMBOL res 304 160 R0 WINDOW 0 56 36 Left 2 WINDOW 3 55 69 Left 2 SYMATTR InstName R1 SYMATTR Value 50 SYMBOL voltage 80 144 R0 WINDOW 0 -146 33 Left 2 WINDOW 3 -264 72 Left 2 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V1 SYMATTR Value SINE(0 5 10Meg) TEXT 552 216 Left 2 !.tran 200n TEXT 448 128 Left 2 ;1N914 Reverse Recovery TEXT 512 176 Left 2 ;JL Feb 11 2019

LTSpice says the snap-off time is circa 0.6ns. What have you measured in real devices?

What other non-SRD device would you choose for a fast step-recovery pulse generator, e.g. in a hobby TDR? (aside from fast logic e.g. CML, TinyLogic, that I'm already aware of)

Clifford Heath.

Am 11.02.19 um 21:08 schrieb John Larkin:

This is a dirty power supply. With it, you can check in a LTspice noise simulation how much dirt you can tolerate on a supply voltage until it starts to impair the performance of your amplifier.

The 60 Ohms resistor provides 1 nV / rt Hz. The VCVS amplifies it to the required level and adds it to clean Vcc. The 100 nV dc source is only there to avoid a simulator crash.

regards, Gerhard

Well, a CML gate or comparator is about perfect. The TEK SD24 used a set of current-steering diodes. There's a patent somewhere.

One can build a fast but ratty TDR and use a fairly simple deconvolution program to make it pretty.

ChesterW and I built a 100-ps-class single-diode TDR with a BOM of about $2, using a pHEMT driven by a line receiver and a very small Schottky diode, all done dead-bug style. 'Tweren't as clean as an SD-24, of course, but it was surprisingly good for what it was.

Cheers

Phil Hobbs