Class i and siblings

dadod

Member
2006-04-18 3:11 pm
Zagreb
Here is quick simulation of an amp with Edmond non switching auto bias OPS.
I appreciate comments and suggestions, specially VAS improvements.
Thanks to Edmond Stuart who is author of this excellent output stage.
I will try to simulate the same IPS with Class i OPS next.
Damir
 

Attachments

  • Non switching EdmondOPS amp-sch.jpg
    Non switching EdmondOPS amp-sch.jpg
    156 KB · Views: 2,662
  • Non switching EdmondOPS amp-sim.jpg
    Non switching EdmondOPS amp-sim.jpg
    255.8 KB · Views: 2,531
  • Non switching EdmondOPS amp-sim2.jpg
    Non switching EdmondOPS amp-sim2.jpg
    196.7 KB · Views: 2,209

dadod

Member
2006-04-18 3:11 pm
Zagreb
Class-i OPS

And now I simulated almost the same IPS connected to the Class i OPS.
Distortion is higher then with Edmond OPS, but it could be improved with better( I suppose) VAS as input resistance of the Class i OPS is highly non linear.
Damir
 

Attachments

  • Non switching Class-iOPS amp-sim2.jpg
    Non switching Class-iOPS amp-sim2.jpg
    216.9 KB · Views: 304
  • Non switching Class-iOPS amp-sim1.jpg
    Non switching Class-iOPS amp-sim1.jpg
    204 KB · Views: 307
  • Non switching Class-iOPS amp-sim.jpg
    Non switching Class-iOPS amp-sim.jpg
    247.3 KB · Views: 2,023
  • Non switching Class-iOPS amp-sch.jpg
    Non switching Class-iOPS amp-sch.jpg
    169.4 KB · Views: 2,166
Class-I

Hi Damir,

Indeed, the input impedance is highly nonlinear. Most likely, that's the cause of much more distortion. See below for the test circuit and its Zi as of function of the output voltage (Actually, it's the output current that alters Zi). Zi varies from 60kOhms to tens of MOhms.

Why improve the VAS (i.e. lower Zo of the VAS) to compensate for deficiencies of the Class-I OPS? Isn't it a waste of time time, as the OPS itself can easily be improved (by means of my AB2 OPS, as shown in your previous post)?

In the next few days, I will show that the nonlinear Zi of class-I is not the only downside of this OPS.

Cheers,
E.
 

Attachments

  • Class-I-Sch-Zi.jpg
    Class-I-Sch-Zi.jpg
    133.8 KB · Views: 1,189
  • Class-I-Zi.jpg
    Class-I-Zi.jpg
    135.9 KB · Views: 1,134
Autobias-II (AB2)

And here for the sake of comparison, the input impedance of the AB2 OPS:
Now, Zi varies between say 1 to 10 MOhms. Needless to say that this variation has far less impact on the distortion, as a typical VAS has far less difficult to handle such a variable load.

edit: Zi simmed at the DC level. Above 10kHz it drops considerably (of course).
See for more info on geometric mean biasing: http://home.tiscali.nl/data.odyssey/AutoBias_II.html

Cheers,
E.
 

Attachments

  • AB2-BJT-Zi-sch.jpg
    AB2-BJT-Zi-sch.jpg
    160.5 KB · Views: 826
  • AB2-BJT-Zi.jpg
    AB2-BJT-Zi.jpg
    118.1 KB · Views: 731
Last edited:

dadod

Member
2006-04-18 3:11 pm
Zagreb
Lower distortion with Class-i OPS.
Separate emitter resistors in Class.i input transistors lowered distortion but it is still ten times worst then with Edmond AB2 OPS. Not sure yet how this change influences thermal stability.
Damir
 

Attachments

  • Non switching Class-iOPS amp-sim3.jpg
    Non switching Class-iOPS amp-sim3.jpg
    199.1 KB · Views: 554
Class-I

Hi Damir,

I see that you have incorporated the filter network (R63, R15 & C23) between the VAS output and the OPS input. However, this filter is only needed if you are using the OPS in a standalone environment, e.g. connected to your tube pre-amp. When embedded in a 'complete' amp, as shown by your post above, you may omit this filter and probably get better distortion figures.

Regarding the 0.33 Ohm emitter resistors (RE), be warned. Since the output impedance (without global NFB), equals RE, any mismatch between the top and bottom RE will affect the performance. Kendall has already warned us: A mismatch of 1% will increase the 2nd harmonic (of 1kHz) to -77dB. Therefore, he recommends the use of precision (0.1%?) high power metal film resistors. I don't think that enough. Also unequal PCB traces and imperfect solder joints may ruin the performance. So I also recommend a Kelvin connection to these resistors.

Cheers,
E.
 

dadod

Member
2006-04-18 3:11 pm
Zagreb
Hi Damir,

I see that you have incorporated the filter network (R63, R15 & C23) between the VAS output and the OPS input. However, this filter is only needed if you are using the OPS in a standalone environment, e.g. connected to your tube pre-amp. When embedded in a 'complete' amp, as shown by your post above, you may omit this filter and probably get better distortion figures.

Regarding the 0.33 Ohm emitter resistors (RE), be warned. Since the output impedance (without global NFB), equals RE, any mismatch between the top and bottom RE will affect the performance. Kendall has already warned us: A mismatch of 1% will increase the 2nd harmonic (of 1kHz) to -77dB. Therefore, he recommends the use of precision (0.1%?) high power metal film resistors. I don't think that enough. Also unequal PCB traces and imperfect solder joints may ruin the performance. So I also recommend a Kelvin connection to these resistors.

Cheers,
E.

Hi Edmond,
Two questions:
1. How critical emitter resistors are in your AB2 OPS, and how to adjust question current for different emitter resistors value(.22R)?
2. I intended to put all in one box, tube pre and OPS. Do I need the filter network in this case(how about AB2 OPS)?
Damir
 
Hi Edmond,
Two questions:
1. How critical emitter resistors are in your AB2 OPS,

Hi Damir.

In the AB2-OPS tight matching of emitter resistors is not critical, as the output impedance depends not RE. The trouble with class-i is that the feedback is taken from the RE's. At idle from both RE's, when sourcing current from the top RE and when sinking current FB is taken from the bottom RE.

In case of the AB2-OPS however, FB is taken directly from the output (as it should be). Therefore, the output impedance is much lower. This is a distinct and important difference.

and how to adjust question current for different emitter resistors value(.22R)?
By increasing the currents through Q9 and Q12 (in post 4), i.e. set R13=R14=100R. BUT I wouldn't do that, as the minimum current through the OP devices comes close to zero. A zero current should be avoided at any time. Set for example R13=R14=91R and you will see the devastation effect.

2. I intended to put all in one box, tube pre and OPS. Do I need the filter network in this case(how about AB2 OPS)?
Damir

Much depends on the output impedance (also at HF) of the pre-amp. It also depends on whether some HF filtering has been done in the pre-amp. The point is that the OPS (both versions) likes to see a low Zo of the preceding stage (in order to avoid instability) and don't like extreme fast transients.

Another point of concern is the DC offset. If the pre-amp is only AC coupled to the OPS (most likely) you have to define the DC operating point of the OPS by means of a suitable resistor at the input to GND. However, if this resistor is too low, the pre-amp doesn't like it (more distortion) and if it is too high you will get more offset. Somewhere in between there must be an optimal value (I hope).

Cheers,
E.
 

dadod

Member
2006-04-18 3:11 pm
Zagreb
Hi Edmond,
Thank you, I think I see now what I am going to do.
As AB2 looks mutch better I will use this one, with my tube preamp, and in all transistors version.
What alternative is for WHAT transistors, matching individual transistors and somehow thermaly connect them?
damir
 
matched transistors

Hi Damir,

Probably you mean THAT transistors (never mind). In principle, you can use any decent low noise small signal tranny (beta > 100; Ft > 100MHz and low internal Rb and Re), but they should be matched, that is, PNP to PNP respectively NPN to NPN. I guess that a delta Vbe of about 3mV (or less) is sufficient.
Regarding thermal coupling, again PNP to PNP resp. NPN to NPN.
Remember, we are dealing here with only 33mV across the emitter sense resistors at idle condition. So any Vbe offset will have a large impact on the quiescent and minimum current of the OPS. This applies to the Class-I as well as to the AB2 OPS.

BTW, Farnell has the THAT300 and THAT320 in stock, though they are a bit expensive, about 9 euro. :sad:

Cheers,
E.
 
Class-I NFB-loop

Now, let's have a look at the NFB loop inside Kendall's Class-I output stage.
In fact here three FB paths: one directly from the output, which is a bit positive (-6dB) and two from the sense resistors, which provide the main and negative FB. At larger output currents, say above 250mA, only one of them is active (either from the top RE or the bottom RE). As a result, the phase and amplitude response of the FB loop depends on the output current. See the graph below.

At idle or low output power conditions:
PM = 60 deg @ 1.5MHz
GM = 6.8dB @ 6.5MHz

And at high output power conditions:
PM = 69 deg @ 2.4MHz
GM = 8.3dB @ 13MHz

I'm not really happy with this variability, as you can't optimize the FB loop for any condition (for small and large signals). In practice this means that you have to tailor the frequency compensation for worst case conditions, that is, at high output currents, as in that case the ULGF is highest. But then the small signal compensation is suboptimal.

BTW, the AB2-OPS doesn't exhibit these quirks.

Cheers,
E.
 

Attachments

  • Kendall2-FB.jpg
    Kendall2-FB.jpg
    120.7 KB · Views: 702

miralin

Member
2010-12-13 7:27 am
Hi Edmond
Basic Auto Bias circuit (Fig.1 in http://home.tiscali.nl/data.odyssey/AutoBias_II.html) has an in-out offset equal to difference between Vbe of pnp and npn transistors. Let me suggest some changes which cannot affect overall work of the circuit.
1.Disconnect collectors of Q5 and Q6 from Out.
2.Connect collectors of Q5 and Q6 to their bases.
We have four transistors connected as diodes. Now negligible I5 and I6 flow thru R1 and R2.
3.Interchange Q3 with Q4 and Q5 with Q6. Now we have diodes on the same places but no offset.
Cheers
Mir
 

dadod

Member
2006-04-18 3:11 pm
Zagreb
AB2 OPS

The FFT at 1kHz shows perfect harmonics distribution up to almost full power.
First one is 1W//8ohm, second one 35W//8ohm and last one 81W//8ohm.
Even at 80W distortion is 1.34ppm.
Damir
 

Attachments

  • Non switching EdmondOPS amp-FFT-1W.jpg
    Non switching EdmondOPS amp-FFT-1W.jpg
    172.5 KB · Views: 672
  • Non switching EdmondOPS amp-FFT-35W.jpg
    Non switching EdmondOPS amp-FFT-35W.jpg
    171.6 KB · Views: 626
  • Non switching EdmondOPS amp-FFT-80W.jpg
    Non switching EdmondOPS amp-FFT-80W.jpg
    179.4 KB · Views: 600
AB-II and siblings

Hi Edmond
No. I mean emitters of Q6 and Q5 connected to S1 and S2, not collectors.
I'm sorry that being a beginner I don't know how to draw and so have to use stiff words instead.

Cheers
Mir

Hi Mir,

Thanks, now I got it. I'm glad you brought this up as I was already waiting for a good opportunity the discuss possible variations on AB-II.
Below you see a number of alternative circuits together with Kendall's Class-I OPS.

The first circuit is the same as in my website. This one is non-optimal and only meant to illustrate the basic principle. The distortion, even at 20Hz, is a bit high: 144 ppm, and the offset is -22mV.

In the second circuit, where the polarity of Q3 and Q4 has been interchanged, the offset dropped to -7.7mV, but the distortion is even higher: 184 ppm.

In the third circuit, that what you meant, right?, also the polarity of Q5 and Q6 has been interchanged. The offset dropped even further: to -2.6mV. (It would be zero in absence of a input current offset.) THD is the same as in Fig.1.

The first three circuits suffer from the same thing: switching currents from Q5 & Q6, which are not canceled out. The effect is clearly illustrated in fig.7: the black curve, which shows the distortion residual.

This effect can be eliminated by crosswise connecting the collectors of Q5 & Q6 to the node A respectively B. Now the THD dropped to 4.3 ppm. That's nice, but the offset voltage is back. BTW, this is a non-issue if the OPS is embedded in an amp with global NFB as well.
Fig.8, black curve, shows the distortion residual (almost gone).

In Fig.5 I've tried to reduce the offset voltage by interchanging the polarity of Q 3 & Q4. Although the offset dropped to -3.9mV, this is not a good idea, as not only the THD rises to 100pm, but also the distribution of drain currents get highly asymmetrical (that is, unequal minimum currents). The reason for above issues is the (inevitable!) mismatch between the PNP and NPN transistors.

And finally Fig.6 shows Kendall's Class-I OPS, that is, with added cascode transistors Q3 & Q6. THD is even lower: 3.3ppm and the offset voltage only 1.5mV. Very impressive, isn't it? Regrettably, also this circuit has some issues: If R1 plus Zo of the preceding stage isn't equal to R2, the THD rises considerably. Also notice that without cascodes, as in the original implementation by Kendall, the performance will be affected. Therefore, (without cascodes) Kendall recommend high Early voltage trannies for Q1, Q2, Q4 & Q5.

Cheers,
E.

PS: I simmed the distortion at very a low frequency on purpose (effectively at the DC level): In order to make the distortion from the bias circuit visible without the contribution from the OP devices.
 
Last edited:

miralin

Member
2010-12-13 7:27 am
AB-II

Hi Edmond
Your post #18-19 is highly informative (I think not only for me), especially about the huge reduction of error due to the crossconnection on Fig.4.
Now let’s return to the “offsetless” Fig.3, which also is the simplest cirquit. Assuming Out grounded, consider e.g. the upper half of the cirquit an amplifier with emitter of Q1 as noninverting input, base of Q1 as inverting input, and Q5Q3 as a nonlinear divider in a local NFB loop. Resulting nonlinearity of this amplifier is presented by a red curve in Fig.7 and is responsible for all errors (black curve).
So almost the only thing determining errors is the nonlinearity of the divider between nodes S1 and A, and the simplest way to change (correct) the pattern of this nonlinearity is to place resistors in parallel or in series with Q5 and/or Q3 (with necessary variations of I3, of course). May be, this can give even better results than the crossconnection, or improve effect of the crossconnection itself.
You have already simmed similar changes, but mostly around Q1 (e.g. in #429 of HEEEELLLPPP and earlier). Why not to do some more experiments?
Cheers
Mir