Re: Re: Re: Re: parametrized open-loop EF dist./harm. vs bias graphs
Hi,
The value of BF may or may not correspond to the actual maximum beta of the device. The best way to explain this is by referring to the Gummel plot in Figure 9 on this page. Beta dropoff at low currents is due to non-ideal base current, while at high currents it's due to non-ideal collector current. If there is no range of Vbe in the Gummel plot for which the ln(Ib) and ln(Ic) plots are parallel, then BF may be much larger than the actual maximum beta.
In the case of the NPN device, if you look at beta vs. Ic, you'll find it keeps increasing up to a very high Ic. This means the non-ideal Ib overlaps the non-ideal Ic quite a bit. If you look at the simulated beta vs. Ic in Figure 16 on this page though, you'll see it matches the measured beta very closely.
JPV said:
Hi,
The value of BF may or may not correspond to the actual maximum beta of the device. The best way to explain this is by referring to the Gummel plot in Figure 9 on this page. Beta dropoff at low currents is due to non-ideal base current, while at high currents it's due to non-ideal collector current. If there is no range of Vbe in the Gummel plot for which the ln(Ib) and ln(Ic) plots are parallel, then BF may be much larger than the actual maximum beta.
In the case of the NPN device, if you look at beta vs. Ic, you'll find it keeps increasing up to a very high Ic. This means the non-ideal Ib overlaps the non-ideal Ic quite a bit. If you look at the simulated beta vs. Ic in Figure 16 on this page though, you'll see it matches the measured beta very closely.
I found, for the output stage measured here:
http://www.diyaudio.com/forums/showthread.php?postid=1297281#post1297281
that CCIF IM distortion, component f1 + f2 is higher for the circuit closed in global NFB than for the circuit without global NFB. Components f2 - f1, 2f2 - f1 and 2f1 - f2 decreased by global NFB. This is only simulation result, not measured.
http://www.diyaudio.com/forums/showthread.php?postid=1297281#post1297281
that CCIF IM distortion, component f1 + f2 is higher for the circuit closed in global NFB than for the circuit without global NFB. Components f2 - f1, 2f2 - f1 and 2f1 - f2 decreased by global NFB. This is only simulation result, not measured.
PMA said:
This is an interesting result. This would be an IM component out at 39 kHz, right? What is the amount of NFB loop gain out there in the design you simulated?
By how much was f1+f2 higher when the loop was closed?
In the design where the loop was open, was the gain flat out to 39 kHz? Was the value of the f1+f2 component the same as the value of the f2-f1 component?
Here's another possible thing to try. Do the CCIF twin tone IM test with frequencies a decade lower, i.e., 1.9 kHz and 2.0 kHz. Then see if the in-band f1+f2 IM component at 3.9 kHz is increased with the application of NFB.
Cheers,
Bob
Bob Cordell said:
Bob,
this is only a simulated result now, though simulation "fits" in case of posted image of CCIF IM.
First, it is 13+14kHz. Loopgain is 37.5 dB at 13kHz and 31.5 dB at 26 kHz.
Hereby the results:
1) no global NFB around output stage
2f1 ..... 26kHz ....... -83.7dB
f1+f2 ... 27kHz ....... -77.2dB
2f2 ..... 28kHz ....... -83.2dB
2) same out. stage inside NFB of same opamp
2f1 .... 26kHz ...... -80.0dB
f1+f2 .. 27kHz ...... -73.7dB
2f2 .... 28kHz ...... -79.4dB
Everything same except for global feedback sense point, speaker terminal in (2) and before output stage in (1). Simulation repeated for 6+7kHz with similar result for 2f frequencies, BUT feedback helped for 3f frequencies!
Re: Re: Re: Re: Re: parametrized open-loop EF dist./harm. vs bias graphs
I understand. Your site is very good. BF is then a parameter which is an extrapolation of the gummel plot and is used with other parameters by spice to model the variation of beta with Ic.
Is this correct?
I believe that it is correct to say that the interesting distortion curves of PMA are coming from three origines: non linear output resistance due to croossover , assymetric beta (npn,pnp) and non linear beta with current.
The second order distortion harmonics should be generated mainly by the assymetric beta but the drop of beta at low currents is mixed with crossover distortion.
Is it possible to modify the models of these thermaltracks to generate three models: one presenting ideal beta(constant and the same for both transistors) that will generate only crossover non linear output resistance like Oliver's paper. Then another one with the actual beta drop but the same for NPN and PNP, then a third one with constant beta's but assymetric(different for npn and pnp).
Running the sim of PMA on these three models will be very instructive.
I am intrigued by the repeated dips looking like sin(x)/x origin.
Jean-Pierre
JPV
andy_c said:
I understand. Your site is very good. BF is then a parameter which is an extrapolation of the gummel plot and is used with other parameters by spice to model the variation of beta with Ic.
Is this correct?
I believe that it is correct to say that the interesting distortion curves of PMA are coming from three origines: non linear output resistance due to croossover , assymetric beta (npn,pnp) and non linear beta with current.
The second order distortion harmonics should be generated mainly by the assymetric beta but the drop of beta at low currents is mixed with crossover distortion.
Is it possible to modify the models of these thermaltracks to generate three models: one presenting ideal beta(constant and the same for both transistors) that will generate only crossover non linear output resistance like Oliver's paper. Then another one with the actual beta drop but the same for NPN and PNP, then a third one with constant beta's but assymetric(different for npn and pnp).
Running the sim of PMA on these three models will be very instructive.
I am intrigued by the repeated dips looking like sin(x)/x origin.
Jean-Pierre
JPV
Before I do any conclusions, I would like to make a short summary of the circuit, measurements and simulations. This information is spread through this and another thread and I understand that it is difficult to follow it.
The output stage explored was shown in
http://www.diyaudio.com/forums/showthread.php?postid=1296735#post1296735
It is biased at about 1.6A and operating in pure class A through the whole amplitude range into 8 ohm. This output stage has much lower distortion compared to class A EF and 2EF output stage, very high input impedance and needs only driving current in 1 - 10uA order.
This output stage, together with HV opamp driver was measured for THD and CCIF IMD 13+14kHz. The output stage was outside opamp FB loop, behind it. The THD amd IMD measured results were copmpared with simulated results and simulation transistor models chosen to give best matching of both results.
A) measurement and simulation results for 13+14kHz CCIF IMD. Measured 2dB under limitation approx.
Frequency Measured Simulated
f2-f1 = 1kHz -92dB -87.5dB
2f1-f2 = 12kHz -82dB -79.5dB
2f2-f1 = 15kHz -82dB -79.3dB
Another spectral components lie below measurement threshold (or in noise floor), which is -120dB. Measured result shown in
http://www.diyaudio.com/forums/showthread.php?postid=1297281#post1297281
The measuring method is calibrated and IMHO gives reliable results. The limit of FFT of measured frequency is 23.5kHz, so 2f1, f1+f2 and 2f2 components were not measured.
----------------------------------------
Following results are only simulated
B) the simulation was done for the same output stage also inside FB loop of the same opamp as in (A).
Loopgain is 37.5 dB at 13kHz and 31.5 dB at 26 kHz.
Hereby the results:
1) no global NFB around output stage
2f1 ..... 26kHz ....... -83.7dB
f1+f2 ... 27kHz ....... -77.2dB
2f2 ..... 28kHz ....... -83.2dB
2) same out. stage inside NFB of same opamp
2f1 .... 26kHz ...... -80.0dB
f1+f2 .. 27kHz ...... -73.7dB
2f2 .... 28kHz ...... -79.4dB
Simulation repeated for 6+7kHz with similar result for 2f frequencies, BUT feedback helped for 3f
frequencies! The gain is flat at f1 + f2, for the openloop output stage. It is flat till at least 200kHz.
---------------------------------------------------
Though the simulation results are interesting, they must be proven by measurements. For this purpose, I will prepare CCIF IMD 6+7kHz method, which would allow me to measure 2f1, f1+f2 and 2f2 components. This would need some time.
Last, the ouput stage works in true class A. There is no dead zone or Gm doubling. The distortion products are IMHO a result of gross signal non-linearity and dissimilarity of n and p halves of the circuit.
The output stage explored was shown in
http://www.diyaudio.com/forums/showthread.php?postid=1296735#post1296735
It is biased at about 1.6A and operating in pure class A through the whole amplitude range into 8 ohm. This output stage has much lower distortion compared to class A EF and 2EF output stage, very high input impedance and needs only driving current in 1 - 10uA order.
This output stage, together with HV opamp driver was measured for THD and CCIF IMD 13+14kHz. The output stage was outside opamp FB loop, behind it. The THD amd IMD measured results were copmpared with simulated results and simulation transistor models chosen to give best matching of both results.
A) measurement and simulation results for 13+14kHz CCIF IMD. Measured 2dB under limitation approx.
Frequency Measured Simulated
f2-f1 = 1kHz -92dB -87.5dB
2f1-f2 = 12kHz -82dB -79.5dB
2f2-f1 = 15kHz -82dB -79.3dB
Another spectral components lie below measurement threshold (or in noise floor), which is -120dB. Measured result shown in
http://www.diyaudio.com/forums/showthread.php?postid=1297281#post1297281
The measuring method is calibrated and IMHO gives reliable results. The limit of FFT of measured frequency is 23.5kHz, so 2f1, f1+f2 and 2f2 components were not measured.
----------------------------------------
Following results are only simulated
B) the simulation was done for the same output stage also inside FB loop of the same opamp as in (A).
Loopgain is 37.5 dB at 13kHz and 31.5 dB at 26 kHz.
Hereby the results:
1) no global NFB around output stage
2f1 ..... 26kHz ....... -83.7dB
f1+f2 ... 27kHz ....... -77.2dB
2f2 ..... 28kHz ....... -83.2dB
2) same out. stage inside NFB of same opamp
2f1 .... 26kHz ...... -80.0dB
f1+f2 .. 27kHz ...... -73.7dB
2f2 .... 28kHz ...... -79.4dB
Simulation repeated for 6+7kHz with similar result for 2f frequencies, BUT feedback helped for 3f
frequencies! The gain is flat at f1 + f2, for the openloop output stage. It is flat till at least 200kHz.
---------------------------------------------------
Though the simulation results are interesting, they must be proven by measurements. For this purpose, I will prepare CCIF IMD 6+7kHz method, which would allow me to measure 2f1, f1+f2 and 2f2 components. This would need some time.
Last, the ouput stage works in true class A. There is no dead zone or Gm doubling. The distortion products are IMHO a result of gross signal non-linearity and dissimilarity of n and p halves of the circuit.
Simulation results at -6dB under clipping :
1) no global NFB around output stage
f2-f1 ..... 1kHz ........ -93dB
2f1-f2 ....13kHz...... -90.8dB
2f2-f1 ....14kHz ..... -90.6dB
2f1 ..... ..26kHz ....... -88.7dB
f1+f2 .... 27kHz ....... -82.4dB
2f2 ....... 28kHz ....... -88.1dB
2) same out. stage inside NFB of same opamp
f2-f1 ..... 1kHz ........-108.3dB
2f1-f2 ....13kHz...... -112.6dB
2f2-f1 ....14kHz ..... -111.4dB
2f1 ........ 26kHz ..... -86 dB
f1+f2 ..... 27kHz ..... -79.7dB
2f2 ........ 28kHz ..... -85.4dB
1) no global NFB around output stage
f2-f1 ..... 1kHz ........ -93dB
2f1-f2 ....13kHz...... -90.8dB
2f2-f1 ....14kHz ..... -90.6dB
2f1 ..... ..26kHz ....... -88.7dB
f1+f2 .... 27kHz ....... -82.4dB
2f2 ....... 28kHz ....... -88.1dB
2) same out. stage inside NFB of same opamp
f2-f1 ..... 1kHz ........-108.3dB
2f1-f2 ....13kHz...... -112.6dB
2f2-f1 ....14kHz ..... -111.4dB
2f1 ........ 26kHz ..... -86 dB
f1+f2 ..... 27kHz ..... -79.7dB
2f2 ........ 28kHz ..... -85.4dB
PMA said:
There is one more possible mechanism for a distortion pattern change - the effect of an output impedance of the opamp VAS. When the loop is closed around the opamp only, that impedance is very low. However when the loop is global, than the opamp output is essentially working open-loop. I wander if that makes a difference?
Cheers
Alex
Hi Alex,
I understand your point. To clarify, the opamp works with 5k1 + 5k1 noinverting network, i.e gain = +2. In (1) the FB point is from opamp output, in (2) from output stage output. Just to define the opamp load somehow, there is always, in both examples, resistor of 39k from opamp output to ground. Yes, the load differs from 10k2//39k to 39k. I may try to equalize it in simulation.
I understand your point. To clarify, the opamp works with 5k1 + 5k1 noinverting network, i.e gain = +2. In (1) the FB point is from opamp output, in (2) from output stage output. Just to define the opamp load somehow, there is always, in both examples, resistor of 39k from opamp output to ground. Yes, the load differs from 10k2//39k to 39k. I may try to equalize it in simulation.
PMA said:
You've got some very interesting results here. Keep up the good work!
If I get a chance, I'll see if I can do some CCIF distortion sims around one of the Class-AB output stages I have previously simulated, so I can see if the NFB closure around it causes an increase in the f1 + f2 products.
Cheers,
Bob
PMA said:
Hi!
it is not the load of the opamp I thought about but the output impedance of the opamp itself. Many opamps would have pretty high open-loop o/p impedance, so the opamp gain could be somewhat modulated by unlinearity of the load. It is rather a resistor in series with the input of the o/p stage you may try to add in a local NFB configuration.
Cheers
Alex
Re: Re: Re: Re: Re: Re: parametrized open-loop EF dist./harm. vs bias graphs
Yes. At low Ic, ln(Ic) vs. Vbe fits a line very well, and at high Ic, ln(Ib) vs. Vbe fits a line very well. If you take the extrapolation of these lines, ln(BF) is the vertical distance between the two on the Gummel plot.
Yes, to make an ideal transistor model in this way is easy. Just specify only Is and BF in the model. To compute Is, pick a collector current and Vbe you want, and solve for Is. Then set BF to the beta you want.
Other combinations are possible. It just involves understanding what the parameters are for the Gummel-Poon model. My web pages show the equations used and have tables translating the variables in the equations to the SPICE parameter names.
Also, another source of distortion to consider is Early effect. If the voltage swing is large, the varying Vcb will modulate beta, in addition to the beta modulation at fixed Vcb and varying Ic.
JPV said:
Yes. At low Ic, ln(Ic) vs. Vbe fits a line very well, and at high Ic, ln(Ib) vs. Vbe fits a line very well. If you take the extrapolation of these lines, ln(BF) is the vertical distance between the two on the Gummel plot.
Yes, to make an ideal transistor model in this way is easy. Just specify only Is and BF in the model. To compute Is, pick a collector current and Vbe you want, and solve for Is. Then set BF to the beta you want.
Other combinations are possible. It just involves understanding what the parameters are for the Gummel-Poon model. My web pages show the equations used and have tables translating the variables in the equations to the SPICE parameter names.
Also, another source of distortion to consider is Early effect. If the voltage swing is large, the varying Vcb will modulate beta, in addition to the beta modulation at fixed Vcb and varying Ic.