Simulating tube amplifiers

Discussion in 'Amateur Radio Amplifiers' started by W8JI, Dec 29, 2011.

Thread Status:
Not open for further replies.
ad: L-HROutlet
ad: l-rl
ad: Left-2
ad: Subscribe
ad: L-MFJ
ad: Left-3
ad: abrind-2
  1. G0HZU

    G0HZU QRZ Member

    My biggest concern about the components used to protect against instability is what happens when
    the instability still occurs despite these components.

    eg I would expect that the anode suppressors could suffer rapid and severe thermal damage if the
    amp hooted strongly up around 120-130MHz. If the resistor in the suppressor blew open circuit then this would mean the loop gain would go UP a lot
    and also the zero phase frequency of the loop would drop. So The hoot would then change down in frequency.

    The impedance at the anode would rise a lot at the new (lower?) hoot frequency because the suppressor is no longer effective and I'd assume large peak voltages could occur at the anode especially if the hoot happened during a high power tune up into a load.

    This lower VHF frequency hoot could put great stress on the 200p caps and also the 1mH
    choke due to high voltages occuring within the grid grounding network in the feedback path.

    Do these parts ever overheat or fail?

    Can anyone advise if suppressors have been tried at the cathode? A simple system model shows a
    marked reduction in loop gain and also a useful shift in phase if something like 75R in parallel with
    120nH is used the cathode.

    However, I don't know enough about the large signal behaviour of the tube to know if the suppressor
    would upset the driver stage performance so maybe someone can comment on this.
    Note: the suppressor won't be effective if there's any shunt components from cathode to chassis
    AFTER the suppressor.
    Last edited: Jan 17, 2012
  2. W1BR

    W1BR Premium Subscriber QRZ Page

    I'm not sure where you are heading with the modeling, since it would seem to be difficult to account for parasitic properties of both the components and the physical layout; but I am enjoying reading about your efforts.

    If you have a library of early Ham Radio Magazines, Mr. Measures ran an article in a late 1980s April issue that showed his modifications
    that included RC bypassing on the grids, and he added resistance in series with the cathode. Try 1986 or 1988?

  3. G0HZU

    G0HZU QRZ Member

    Hi Pete, because instability occurs at a single frequency you can simplify the early models to get a better understanding of the root cause of that particular instability event.

    Eg the entire tank circuit complete with parasitics can be replaced by a single series resistor and a single series reactance at this ONE frequency (in this case inductive reactance)

    The same applies to the grid decoupling network. Obviously this isn't the 'same' as soon as you look either side of this spot frequency but it is a good method to start with.

    Adding resistance in series in the cathode is one way to reduce VHF loop gain but if you do it with a parallel RL network eg 75R in parallel with 120nH you get about 40R in series equivalent at VHF but only 6R at 29MHz. So the loop gain reduces a lot more at VHF plus you get the phase shift in the inductance and Cgc that shifts the zero phase around the loop away from the peak in VHF loop gain.

    I'll try a web search for those articles in HR :)
  4. G0HZU

    G0HZU QRZ Member

    I've quickly put together a series of very basic system models on Genesys that show how the K factor gets affected as the amplifier model 'grows' from a basic model with perfect gate grounding to one with 0.07pF Cdc.Then gate inductance is added in circuit 3 and you can see the stability factor crash way below 1. Then the effect of adding suppressors is shown.

    (Although the amplifier model uses a JFET symbol it is just a very basic maths model of an amplifier. It is an AC model so no need to include biasing or DC blocking etc)

    Then in circuit 6 you can see how having tube socket with inductance values for the socket and the copper tape added. You can see the stability gets much worse.

    Then finally in circuit 7 you see the benefit of the 200p caps even though they have 10nH 'extra' leg inductance.For unconditional stabilty K has to be >1. If it is >1 then the model doesn't care about what source and load termination you attach to the amplifier. It should still be stable. However, if it dips below 1 then the amp can go unstable with certain source and load impedances.

    See attached word doc (assuming word docs can be attached)

    View attachment stab2.doc
    Last edited: Jan 18, 2012
  5. G0HZU

    G0HZU QRZ Member

    So what have we learned so far?


    A model of the output network would be fairly simple to create. Use a modern(ish) VNA and export 1 or 2 port S parameter data of the whole tuning tank circuit.

    Capturing data models of complex (unknown black box) RF circuits has been possible for decades and takes very little time these days.

    I've been doing it for over 20yrs but it goes back much further than this. So that weak argument is based on unawareness of the capability of modern RF tools.

    You don't NEED to generate a spice model of the tube. If you assume the instability mode begins as small signal instability you can analyse the
    circuit as a linear system. You could do what I did and use a very basic maths model for the tube or, it is possible (if you are brave) to take live S parameter data of the tube using test equipment and create a pretty good small signal model of the tube in its socket.

    This would have to be done in a controlled (safe!) environment like an RF lab with great care to prevent electric shock and would require a VNA that can work with an external coupler/pads rated to handle the possibility of the amplifier going unstable (to prevent damage to the coupler or VNA)
    But you could take 2 port data at the tube cathode and anode (without the tuning tank) and create a useful small signal amplifier model. This model would capture all parasitics due to layout and the socket and the internals of the tube.

    This data could then be used to analyse the effectiveness of suppressors and you could also generate tube models for different makes of tube.

    Obviously, this couldn't be done safely unless carefully planned and carried out in an RF lab but you could get this 2 port data if you really wanted it.

    I do get the impression that many of the people on this thread aren't really aware of modern analysis methods (eg K and B1 factors for stability analysis) and the ability of modern RF test gear to generate accurate models of complex linear circuits (including amplifiers).

    i.e. some people are still stuck in the 1960s or 1970s.
  6. G0HZU

    G0HZU QRZ Member

    To give an example of modelling using S parameter data you can visit the MiniCircuits website and download datasheets etc for their amplifiers.

    The example above is the GALI-51. They already supply S parameter data for this amplifier to allow designers to models circuits with this device.

    But it is very simple to generate your own S parameter data file using a VNA.
    eg I generated S11, S21, S12, S22 data here by putting one of their GALI 51 eval PCBs on a VNA and exported my own S Parameter data for the eval PCB. This models the whole PCB including layout effects and effects from the bias components and blocking capacitors.

    The advantage of generating your own S parameter data is you can do it at different bias settings or at different temperatures or even on different
    PCB dielectric. This allows more accurate modelling of your own design including PCB layout effects.

    It isn't just restricted to amplifiers. You can take S parameter data of many things such as aerials. eg I recently took S data of my aerial in the
    garden (also took S parameter data of the feeder) and used this to design a simple and efficient ATU on a computer.

    The aerial model includes all stray effects from other nearby objects in my garden.

    You could even take a pretty accurate model of it when it was wet or if a tree was removed from the garden or if birds sit on the aerial.
    I would guess that many people on here would have initially said that such a computer model was impossible to obtain LOL!
    Last edited: Jan 24, 2012
  7. G0HZU

    G0HZU QRZ Member

    Now is probably a good time to show a practical (relevant?) example of the power of using modern RF tools to analyse 'black box' circuits.

    The doc below outlines the results of me using a VNA to take s parameter data from the JFET amplifier I built earlier.

    It shows how this data can be used on a computer to accurately predict where the amplifier will go unstable if you attach a certain load impedance to it.

    The whole point of the doc is to show that you don't need to know ANYTHING about the amplifier device itself or any circuit strays because this all gets captured by the VNA. So the approach outlined below could have been used to model a tube amplifer. (but you would have to use a very rugged outboard coupler plus meaty attenuators to protect the VNA from possible overload damage)

    See attached doc

    Attached Files:

    Last edited: Jan 24, 2012
  8. G0HZU

    G0HZU QRZ Member

    Can you now agree that with modern tools we can quickly capture accurate linear models of VERY complex circuits (including amplifiers in sockets)?

    Everything you have listed above can be overcome using modern tools to capture small signal data models of active or passive networks even if they are fitted to poor quality sockets with less than ideal connections.

    The VNA doesn't care. It will produce a linear (small signal) model of the WHOLE NETWORK. Warts and all... :)

    This can be imported into a linear simulator on a computer as a 2 port model of the network. As I said earlier, the best way would be to break the system up into sections and produce 1 or two port models of each section with a VNA and cascade them into a system simulation on a PC. This way the effect of small changes between each section can be analysed quickly.

    That's how modern RF design or analysis is done. The system model will never be quite as good as the real thing but usually it's good enough to prove USEFUL and can often help the designer discover things about the system stability that aren't obvious to a human observer who merely tinkers with the system trying to 'fish' for instability in a search for better understanding.

    I've been using modern analysis tools for over 20 years as an RF designer and in my opinion if someone says you can't model an RF system then maybe it means they don't understand the system as much as they think they do. Or maybe they don't realise how powerful (and accessible) modern RF modelling techniques have become.
    Last edited: Jan 26, 2012
Thread Status:
Not open for further replies.

Share This Page

ad: elecraft