Has anyone seen Rush pairs used this way? For a CFA with true differential inputs

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi Ian
That looks like fun, perhaps a bit optimistic with output transistors perfectly matched for BF and, less plausibly, for VAF.
I am curious about the stability, the Tian probe placement in the thumbnail may not be fully informative.
What does the Return Ratio look like?

Best wishes
David
See attached.

Perfectly matched transistors are used to see what causes the remaining distortions. Most of the distortion seems to be nonlinear loading of the VAS by the output stage. Adding buffering could reduce distortion. I haven't tried that yet.

Cheers
 

Attachments

  • Rush+VAS-mirror-OPS-prb_05-Jul-2018.png
    Rush+VAS-mirror-OPS-prb_05-Jul-2018.png
    20.4 KB · Views: 301
See attached...

Hi Bimo!
Did you miss Ian's outputstage? Probably the goal of this design is not in the input stage, and VAS,
connecting it to double EF, like your sch/sim showed, will not give the benefit.
You have lots of HF harmonics from switching, which are neglible in Ian's circuit.

...from Ian's web site and perhaps look again the earlier one you are referring to above for a comparison of their output voltage levels?

Like for like seem more or less similar at ~20+dB?

Cheers, ian
 

Attachments

  • 7_Rush+VAS-mirror-OPS-100W-20kHz-4ppm-Tj27C-67C_05-Jul-2018.png
    7_Rush+VAS-mirror-OPS-100W-20kHz-4ppm-Tj27C-67C_05-Jul-2018.png
    38.2 KB · Views: 255
OK thanks...

Ian published it in the post I linked
I sent the results of sim the same circuit but with higher voltage output

... this is interesting.

It seems that your results which were produced under even more demanding conditions are somewhat better than Ian's own published data obtained at a lower output level?

I wonder what could account for that given the same circuit, components and loading etc?

Cheers, ian
 
Hi Bimo!
Did you miss Ian's outputstage? Probably the goal of this design is not in the input stage, and VAS,
connecting it to double EF, like your sch/sim showed, will not give the benefit.
You have lots of HF harmonics from switching, which are neglible in Ian's circuit.

But the title of this thread is about input stage :)
May be I will change the output stage.
 
Perfectly matched transistors are used to see what causes...

OK, now it's clear, that's more or less what I do too but I usually reset to the untouched models before I show the simulated results, for both simplicity and realism.

Thanks for the Tian plot, the probe is placed as shown in post #12?
I think this may be inaccurate since it doesn't capture the feedback from NFB via C5.
I don't know how much difference there will be in your particular circuit but I think it should be placed at NFB before it splits.
Very interested in this, looks like the basis for a rail-to-rail OPS.

Best wishes
David
 
Last edited:
Sorry, I should have reverted to realistic Beta and VAS mismatch before posting.

Re C5 and Tian probe plots.
I haven't read the Tian paper(s) on his probe but after some thought I figure that C5 can be left as is in my circuit. My reasoning is that C5 alters the forward path transfer function (G), and it's effect will be accounted for as part of the combined feedback transfer function (1+G*H) with the probe. I looked again at Ed Cherry's analysis of his NDFL's and his equivalents to C5's get absorbed in his math's analysis into the forward 'G' parts.

I tried as you suggested and put C5 to the other side of the probe and got less gain and it improved the GM by 2dB and PM by 4 degrees. So as I used it it gives slightly lower (worse) readings than the way you suggest.

Has there been any discussion on diyAudio.com previously about how the likes of C5 should be wired with the Tian probe?

BTW If you have read some of my posts, articles you may know that I am not into high feedback and ultra-low distortion amps. I did this as a lead-up to compensation for AndriyOL's 'SSAHP' thread and have just posted some OITPC mods here One of the Top Solid-State CFA amp design.

I got nearly an order of magnitude lower THD at 20kHz than single pole Lead-Lag plus Miller. But I couldn't get the PM and GM large like Damir did in his amps. There maybe some trick I haven't found yet for this.

Cheers
 
Sorry, I should have reverted to realistic Beta and VAS mismatch

No problem, there are some classic "tricky" circuits that work in simulation with perfectly matched transistors but not in the real world so I was a bit wary, but now I understand.

Re C5 and Tian probe plots.
C5 alters the forward path transfer function (G), and it's effect will be accounted for as part of the combined feedback transfer function (1+G*H) with the probe.

Yes, this is true if there is no forward transmission thru feedback loops or other non-ideal behavior, so it all fits the ideal model.
Usually this is very close to accurate but I have been surprised once or twice when there was non-ideal behavior, so I don't entirely trust my expectations.

I looked again at Ed Cherry's analysis of his NDFL's and his equivalents to C5's get absorbed in his math's analysis into the forward 'G' parts.

Yes! that was one of the first papers that made me really think about this, did you look at the JAES paper or the more detailed paper in Circuit Theory and Applications?
The JAES paper (and even more so the Electronics Today articles) tends to skate over the implications of that "absorption".
He later wrote (admitted?) that it tended to obscure the conditional stability of the resultant system.

I tried ... C5 to the other side of the probe and ...it improved the GM by 2dB and PM by 4 degrees.

Excellent, as I said it usually doesn't make much difference but nice to be sure.

Has there been any discussion on diyAudio.com previously about how the likes of C5 should be wired with the Tian probe?

There was a protracted discussion in the context of Transitional Miller Compensation.
Even some of the experts were mislead but eventually a consensus was reached, mainly driven by the esteemed "JCX".
This was mostly before I joined DIYAudio but I have had discussions on the subject with various people, Damir especially likes to make complicated compensation schemes that tax my ability to analyse;)
Now Damir and I are pretty much in accord I think.

...like Damir did in his amps. There maybe some trick I haven't found yet for this.

Damir is pretty sharp!
I see the "trick" as the realization that Bode's calculation of phase conservation should be called Bode's Iron Law;).
If you want extra phase in one area it has to come from somewhere else and the way to move it is to move the gain.
So I see it as an exercise to shape the gain to move the phase where we need it, can't ever beat Minimum Phase but one can do worse if not careful.

I was able to achieve this after some work >Middlebrook GFT probe?<

Best wishes
David
 
Last edited:
I compared CFA with and without Rush cascode input in simulation. Both using same VAS and same output stage. I used two pole compensation like Damir's with slightly different implementation.

Rush cascode input have slightly higher THD at 20kHz about 2x. But DC Offset like Blameless topology, not much change with temperature. I think it save without DC servo or without DC Offset adjustment. Feedback resistor can be higher because high impedance at feedback input. Slew rate did not affect by resistor feedback value (?).

PM and GM.
 

Attachments

  • Untitled2.png
    Untitled2.png
    85.5 KB · Views: 354
Rush+VAS versus LTP+VAS

I compared CFA with and without Rush cascode input in simulation. Both using same VAS and same output stage. I used two pole compensation like Damir's with slightly different implementation.

Rush cascode input have slightly higher THD at 20kHz about 2x. But DC Offset like Blameless topology, not much change with temperature. I think it save without DC servo or without DC Offset adjustment. Feedback resistor can be higher because high impedance at feedback input. Slew rate did not affect by resistor feedback value (?).
Thanks Bimo. I also fund a 2x gm difference. This is most likely due to 4 b-e junctions per input volt compared to 2 b-e junctions for a diamond or a LTP.

But when two mirror are used cross-fed from the input collectors the gain is restored. As I used in post 12 (Q19,Q20 & Q20,Q21).

keantoken kindly did a comparison for me of the standard LTP with the Rush + mirrors. keantoken said I could share it. See attached files. I have tweaked them for best GM's and PM's.

To compare like-with-like the LTP version has the same transistors, the same degeneration resistors and feedback resistors, and same currents in the input transistors and the VAS. The same effective Miller capacitors are used (the Rush has the Miller capacitance split into two).

The differences are the LTP has a 'zero' resistor in series with the Miller capacitor and the Rush has a lead capacitor in parallel with the VAS degeneration resistor -- they are different to maximize the GM and PM of each version. They differ slightly in tweak methods. The Rush uses a 22k resistor across the mirrors that drive the VAS and this resistor is chosen so the LF gain is the same as the LTP version (145dB) for a like-like comparison. This shunt resistor reduces the sensitivity to R2,R3 and mirror matching tolerances for the VAS idle current.

These simulations show the Rush has almost the same specifications as the LTP equivalent. THD at 20kHz and 4V rms is -118dB for the LTP and -126dB for the Rush. The effective VAS load is 10k which is lower than most output stages.

But the Rush version uses 22 transistors and the LTP uses 12 transistors. Is there enough benefit to justify the extra parts? Maybe...

The output offset voltages are 57mV for the LTP and 6mV for the Rush. So the Rush version cancels most of the offset voltage without trimming. If you don't want a trimpot for the DC null and don't want a servo then this seems to be an option.

But can this low offset be achieved in practice with typical transistor parameter spreads? At present the PNP and NPN mirror pairs need to be very closely matched. Maybe Rush mirrors can overcome this as well?
 

Attachments

  • KT_NTP-vs-LTP-psrr.asc
    15.4 KB · Views: 49
  • KT_NTP-vs-LTP-ac.asc
    15.3 KB · Views: 53
  • KT_NTP-vs-LTP-sine.asc
    15.3 KB · Views: 56
...Yes! that was one of the first papers that made me really think about this, did you look at the JAES paper or the more detailed paper in Circuit Theory and Applications? ...
It was CTA: Cherry Ed, ‘A new result in negative-feedback theory, and its application to audio power amplifiers’, IEE Circuit Theory and Applications, Vol 6, 1978, p265-288. I have it on MyDrive here.

Also on MyDrive here: Cherry Ed, 'Feedback amplifier configurations', IEEE Proc. Ccts Dev Syst Vol 147 No 6 Dec 2000, p334-346. Generic feedback configurations systematic analysis. He analyses CFA's CFB and compares them to VFB. Interesting paper.

BTW these are located in IansArticles a subfolder in IansCubelawAmps (see my sign for link).

Thanks for your thoughtful information.
 
To compare like-with-like the LTP version has the same transistors, the same degeneration resistors and feedback resistors, and same currents in the input transistors and the VAS. The same effective Miller capacitors are used (the Rush has the Miller capacitance split into two).

I am not interested to compare LTP version with CFA's Rush Coscode input.
Comparison of LTP and regular CFA is done at one thread (long thread) here and Bonsai made good article of it: http://hifisonix.com/wordpress/wp-content/uploads/2014/01/CFA-vs-classic-Lin-VFA-topology.pdf

But I am interested to compare regular CFA with CFA's Rush Cascode input. Maybe we can not call it "CFA" for Rush Cascode input :rolleyes:

How DC and AC behavior? How to calculate slew rate? This is important because slew rate is the main advantage of CFA topology. Although VFA topology can have very high slew rate but I think it can not as simple as CFA topology.
 
Rush VFA versus Rush CFA

I am not interested to compare LTP version with CFA's Rush Coscode input...

But I am interested to compare regular CFA with CFA's Rush Cascode input. Maybe we can not call it "CFA" for Rush Cascode input :rolleyes:...
I'm not sure what you mean by "regular CFA". An example might help me.

I have now compared the VFB Rush input stage (like the one for keantoken earlier) with a CFB Rush input stage. To convert it to CFB I connect the inverting buffer transistors to trans-diodes and lower the feedback resistors by a factor of 20 (now 500R and 12R previously VFB 5k and 250R). Then trimming R6,R7 for the original VAS current of 12mA. I also changed the OITPC capacitor (C9) to after the Tian voltage source as recommended by Dave Zan in posts earlier because large errors arise with the CFA version. I added a HF pole to the output stage buffer which is essential for tuning OITPC.

Keeping all other things the same the CFB version has half the loop gain and twice the THD of the VFB version. This is due to the 12 ohm CFB resistor (R3) which degenerates the inverting input by 3dB that the VFA version doesn't have. 12 ohms is about the lowest practical value due to the power rating of R4 to be under about 3W (eg two 2W MF or four 1W MF).

This shows CFA distortion in practical power amps can be slightly higher than a buffered CFA (a.ka. VFA) due to the degeneration effect of the resistor on the inverting input that the VFA version doesn't have. This is when all other variables are matched like I have tried my best to do.

There are slight differences between the CFB and VFB versions when compensating. The VFB version benefits from capacitance across the feedback resistor whereas the CFB version is best with bypass capacitors across the VAS emitter resistors.

These CFB and VFB input stages were then used with my Class-AB+C output stage. The CFB version also had twice the THD of the VFB version (FFT plots attached).

It appears the buffered Rush input stage (Post 3) is better than the alternative unbuffered Rush input stage (Post 1). It appears the slight difference in buffer transit times become insignificant in the context of audio power amplifiers since a small capacitor can be added across the feedback resistor (eg 5pF//5k for the attached VFA) to offset this.
A more recent addition to the Rush input stage was the current mirrors before the VAS's with a bias stablizing shunt resistor (eg 22k) across the mirrors (see attached circuits). It first appeared in Post 12 here.

Cheers
 

Attachments

  • Rush-CFB+VAS-mirror-prb_19-Jul-2018.asc
    13.9 KB · Views: 69
  • Rush-VFB+VAS-mirror-prb_19-Jul-2018.asc
    13.9 KB · Views: 56
  • Rush-CFB+VAS-mirror_19-Jul-2018.png
    Rush-CFB+VAS-mirror_19-Jul-2018.png
    27.7 KB · Views: 311
  • Rush-VFB+VAS-mirror_19-Jul-2018.png
    Rush-VFB+VAS-mirror_19-Jul-2018.png
    26.9 KB · Views: 324
  • Rush-CFB+VAS-mirror-OPS-120ppb-1W-20kHz_19-Jul-2018.png
    Rush-CFB+VAS-mirror-OPS-120ppb-1W-20kHz_19-Jul-2018.png
    53.2 KB · Views: 309
  • Rush-VFB+VAS-mirror-OPS-70ppb-1W-20kHz_19-Jul-2018.png
    Rush-VFB+VAS-mirror-OPS-70ppb-1W-20kHz_19-Jul-2018.png
    14.9 KB · Views: 286
Last edited:
Hi
I don't believe the first schematic in post-1 uses Ruish pairs; rather, the input device emitters are simply tied to a voltage reference.

Look at Cherry's NDFL amp.

I used Rush pairs before I'd heard of them in an amp I built in 1985. The amp worked but had problems and I didn't have any decent test equipment to figure it out back then.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.