Its been a week since I am trying to simulate in LTSpice a simple amplifier using a transformer with CFB output stage. My outcomes so far are inconclusive as far as the output level and distortion is concerned. In general it seems that lower levels of CFB produce more power and less distortion which is the opposite of what I was expecting. Moreover, what is a normal level for a CFB amplifier I found citations for 12,5%, in another article about Williamson 100w I found 14db. I am really confused, any help?
Increasing the CFB should improve the distortion figure, but makes life hard for the driver. I would check the distortion from the driver as well. Improvements in one may be cancelled by problems in the other.
The higher the CFB is, the less +Ub is "available" for output tubes and therefore less output power.
Can you post your simulation (.asc) ?
If you put a AC current source across R11 remove the Global FB and plot the speaker voltage you can get a bode plot.
LTSpice CFB simulation
Hi. more than a year ago I did a lot of work simulating a CFB output stage (amongst many others) to see what I might attach to tubelab's Universal Driver Board. As smoking-amp noted you need suitable drivers!
Attached is a zip file containing my model file and the includes. There's a lot I can take out if the model that doesn't impact it's performance but I never got around to it.
Also included is an Excel spreadsheet. I wrote a macro to extract all the data from the LTSpice error log which I could then chart. Don't worry about the links it won't be able to load.
CFB performance was very promising.
If the current crazy drive output stage I have setup on the bench doens't work out I may well go back to CFB for the final design.
I have some other files with different transformers like the Van-der-Veen toroidal one.
When I load your file LTSpice complains it can't find symbols for tetrode_dd and triode_dd.
Hi. more than a year ago I did a lot of work simulating a CFB output stage (amongst many others) to see what I might attach to tubelab's Universal Driver Board. As smoking-amp noted you need suitable drivers!
Attached is a zip file containing my model file and the includes. There's a lot I can take out if the model that doesn't impact it's performance but I never got around to it.
Also included is an Excel spreadsheet. I wrote a macro to extract all the data from the LTSpice error log which I could then chart. Don't worry about the links it won't be able to load.
CFB performance was very promising.
If the current crazy drive output stage I have setup on the bench doens't work out I may well go back to CFB for the final design.
I have some other files with different transformers like the Van-der-Veen toroidal one.
When I load your file LTSpice complains it can't find symbols for tetrode_dd and triode_dd.
dch53, the NDF04N60Z is missing for me.
I'm also looking at CFB for an amplifier design somewhat similar to the one dch53 just posted. My results also showed it is a very promising addition distortion-wise, as well. I'm also planning on using MOSFET drivers.
My understanding was that GNFB was difficult to model in Spice and that the values were best determined by measurement after the amp was built. With the multiple coupling capacitors, there are definitely going to be phase shifts and what baudouin0 said about creating a Bode plot may be necessary. A Bode plot will show you the phase change in relation to frequency.
Due to the phase shifts in the amplifier, is it possible that you are creating a positive feedback loop, rather than a negative one? I would first leave out the GNFB for testing the levels of CFB, then add it in later, if needed.
I'm also looking at CFB for an amplifier design somewhat similar to the one dch53 just posted. My results also showed it is a very promising addition distortion-wise, as well. I'm also planning on using MOSFET drivers.
My understanding was that GNFB was difficult to model in Spice and that the values were best determined by measurement after the amp was built. With the multiple coupling capacitors, there are definitely going to be phase shifts and what baudouin0 said about creating a Bode plot may be necessary. A Bode plot will show you the phase change in relation to frequency.
Due to the phase shifts in the amplifier, is it possible that you are creating a positive feedback loop, rather than a negative one? I would first leave out the GNFB for testing the levels of CFB, then add it in later, if needed.
I had a lot of success simulating the global negative feedback when I built an amp. The results were pretty much the same as the real thing.
All components and valves are pretty accurate. However its the output transformer that is difficult. Basically I had to measure the transformer up on the bench measuring resistance turns ratio primary inductance. Then I did some gain and phase plots with the correct transformer loads. I then did the same thing in LT spice and adjusted the coupling factor and primary capacitance until I got the same results.
Mine was a hammond 1650T but that not much use to you.
You may find the coupling factor is wrong in your model. Quite often its to low! Sometimes they get calculated from the quoted frequency response of the web site. Quite often this may say 20Hz-30KHz, but in fact the frequency reponse may actually be 5Hz-80KHz. So I think you will need accurate gain and phase data for your transformer.
If your not sure about bode plots there's plenty of reading. For LT spice there's no bode plot as such its just a matter of breaking the negative feedback and adding a AC source on one side and measuring the voltage on the other. The only thing is to make sure that the AC source does not affect the circuit impedance. So if you have a cathode resistor you would need to make sure a voltage source has a much higher injection impedance than the cathode or use a current source. Hope that make sense - else somebody can better explain.
All components and valves are pretty accurate. However its the output transformer that is difficult. Basically I had to measure the transformer up on the bench measuring resistance turns ratio primary inductance. Then I did some gain and phase plots with the correct transformer loads. I then did the same thing in LT spice and adjusted the coupling factor and primary capacitance until I got the same results.
Mine was a hammond 1650T but that not much use to you.
You may find the coupling factor is wrong in your model. Quite often its to low! Sometimes they get calculated from the quoted frequency response of the web site. Quite often this may say 20Hz-30KHz, but in fact the frequency reponse may actually be 5Hz-80KHz. So I think you will need accurate gain and phase data for your transformer.
If your not sure about bode plots there's plenty of reading. For LT spice there's no bode plot as such its just a matter of breaking the negative feedback and adding a AC source on one side and measuring the voltage on the other. The only thing is to make sure that the AC source does not affect the circuit impedance. So if you have a cathode resistor you would need to make sure a voltage source has a much higher injection impedance than the cathode or use a current source. Hope that make sense - else somebody can better explain.
Last edited:
I was afraid someone would say that. Anyway, I found the 2 files are they're attached.
I would be leaving out the GNFB for testing. When setting up the operating points tubelab's UDB the idea is to minimise the open loop distortion.
I'd like to do all that when I finally get around to choosing a configuration and a transformer.
I did it some time ago, with improved Williamson and floating paraphase. Additionally, I built prototypes of both. Optimal CFB (from my perspective) is around 12%.
For reference, Sansui AU111 have output transformer with 10.5% CFB.