Is there anybody built a non feedback amplifier??

[AKSA]

Steven,

Forgive me; I'm already peforming ritual amputation on your circuit!

Your CFP diff input stage is terrific, like it a lot. No change.

However, your voltage amplifier is a variation on Lender's circuit, whereby he uses a single pnp (looking at the positive rail) in the present position with its base driven NOT from the emitter of a follower, but directly from the left side rail load of the diff pair.

With appreciable current flowing through the diff pair, say 1mA per side, 2mA stage, I can't see why there's a need for the emitter follower. Of course, knitting together the two pairs of diff stages would need to be done with a much higher value of resistor, but I'd seek to use just one diff pair anyway, and stick with Lender's configuration for VAS. Any thoughts on this?

BTW, no requirement for lag compensation is a terrific step forward, a direct result of no negative feedback. I'm presuming that loop gain is controlled by the resistive loading to ground on the VAS. Correct?

It is still not clear to me how all outputs never switch off. I suspect the answer lies in the VAS itself. Can you please explain?

Cheers,

Hugh

 

[PMA]

Hugh,

power followers are starting projects for the beginners, ZEN - like category of circuits. As you properly stated we should distinguish between simple easy-to-build circuits and continuous attempt to get the best sonic results, that certainly does not lead through these very simple circuits .

The image shown by Steven is named "hawksford.gif" (have a look at the image properties) and Steven surely knows why he named it this way. I sincerely suggest to you to study articles of Malcolm Hawksford, patents of Bruce Candy and DC102 topology of Erno Borbely. You will see that all of these inspirations are more than 20 years old and very similar to the circuit shown. All of us continuously learn and try to assembly the best parts of individual topologies to the best circuit (sonic) result, sometimes enlightenned by lucky fortune and coming with circuits of our own .

 

[AKSA]

Pavel,

I shall probably decline your suggestion to further study audio amplifier theory; I covered most of this ground some years ago and now wish to rest on my laurels....

Thank you, however, for giving me your time and explaining with such patience. You are a truly intelligent, tolerant man.

Hugh

 

[PMA]

Hugh,

it's a pleasure to be here.

 

[Steven]

Interesting circuit, Steven. Yes, it is important to find novel ways to steer the output so that it has no DC offset. Personally, I believe in servos. BUT, it seems important to me that the servo is only significantly active at VERY LOW frequencies, and not as a low frequency roll-off. Years ago, I actually tried to make a servo also act as a high pass filter (ie 50Hz). This was not so good. It appears to be better to just use a quality cap to do the same thing.
Usually, it is best to minimize 'global' feedback. Probably, feedback pairs are better, but not perfect. In any case, the most linear circuit, (without feedback) is usually best.

Thanks John,

In this case the servo has a time constant of 1 sec and its "swing" via the current sources to the actual (DC) output of the amplifier is limited. Bandwidth measurements of the complete amplifier show a low frequency roll off below 1Hz.

I can imagine that using a servo as normal HPF within the audio band reveals the sonic characteristics of the servo circuit itself. And mostly that is just a simple opamp chosen for low offset and especially for low input bias currents (because that allows for a high input resistor and moderately sized integration capacitor), not a dedicated audio opamp. In the amplifier above, a TL061 has been used (in 1983).

Steven

 

[Steven]

... Steven, Pavel, are you familiar with Peter Blomley's non-switching Class AB of the early seventies? There's another to consider.

Yes, I'm familiar with Blomley's class B. I have a copy from his article in the Wireless World in 1971. Actually it is not completely non-switching. The ouput stage is non-switching, but the class B switching has been moved to a previous stage, where it can be accomplished with smaller high speed switching transistors. In 1984 I joined a small 'audiots' club with friends listening to each others amplifiers, loudspeakers and so on (most of it DIY) and one of them designed and built an amplifier based on the Blomley topology. It sounded very good, surprisingly good. I had thought that the hard switching stage would have had some nasty audible artefacts, but they were not there. He published his design in June 1992 in the EW+WW.

Steven

 

[Tube_Dude]

Steven

A very nice circuit...i agree with you about the qualitys for a op amp used in a servo...and in the John opinion that the servo must be only active at VERY low frequencyes for less interference in sound quality...that way the "sound" quality of the op amp come out from the equation...

Only a question...why you don't use voltage feedback from the oposites sides of the zenner to the inverting input using two resistors...one from each side of the Zenner...Just curiosity!!

 

[Steven]

...
However, your voltage amplifier is a variation on Lender's circuit, whereby he uses a single pnp (looking at the positive rail) in the present position with its base driven NOT from the emitter of a follower, but directly from the left side rail load of the diff pair.

With appreciable current flowing through the diff pair, say 1mA per side, 2mA stage, I can't see why there's a need for the emitter follower. Of course, knitting together the two pairs of diff stages would need to be done with a much higher value of resistor, but I'd seek to use just one diff pair anyway, and stick with Lender's configuration for VAS. Any thoughts on this?

BTW, no requirement for lag compensation is a terrific step forward, a direct result of no negative feedback. I'm presuming that loop gain is controlled by the resistive loading to ground on the VAS. Correct?

It is still not clear to me how all outputs never switch off. I suspect the answer lies in the VAS itself. Can you please explain?

Cheers,

Hugh

Hugh,

We tried a number of different topologies for the VAS stage at that time and the one chosen was both simple, had good linearity and bandwidth. Cascoding was not required with this one. But, maybe you're right, the current in the emitter follower is only twice as high as in the left half of the diff pair. On the other hand, because the emitter follower takes care of Vbe cancellation in the VAS stage, the load resistor of the left half of the diff pair to the positive rail (I should have kept the resistor labels in the circuit) should have been increased to get the same VAS current, and this would have increased the gain, and this would require a lower VAS load resistor to keep the amplifier gain equal (open loop), and this would require a higher VAS current to give the same swing, etc. etc. Or a diode should have been inserted in series with the load resistor of the left half of the diff pair to the positive rail to compensate for the VAS Vbe.

Yes, the gain of the amplifier is determined by the VAS load resistors, one at each side of the zener diode, in combination with the diff amp load resistors and the diff amp emitter resistors. Your assumption is correct.

The outputs never switch off completely because the emitter resistors of the output transistors are really small and because of the action of the error feedback amplifier that also takes care of the biasing. It's not the VAS that takes care of not switching; the VAS is just simple class A.

Steven

 

[Steven]

...
The image shown by Steven is named "hawksford.gif" (have a look at the image properties) and Steven surely knows why he named it this way.
...

Pavel,

It's not a secret that could be discovered by looking at the image properties. I also mentioned it in writing in the same post. The output stage uses Malcolm J. Hawksford's error correction circuit. Details about this can be found in the JAES, Vol. 29, January/February 1981, p. 27-31. Title: Distortion correction in audio power amplifiers.

I always wondered why so few amplifers use this error correction scheme. I only saw two other designs, one from Robert Cordell, and one from Peter van Willenswaard. It's simple and works very well, although, I admit, it's not very intuitive. As far as I know it is not covered by patents.

Steven

 

[Steven]

...
Only a question...why you don't use voltage feedback from the oposites sides of the zenner to the inverting input using two resistors...one from each side of the Zenner...Just curiosity!!

Jorge,

That could have been done, but we wanted to use only local feedback and no multi stage feedback. Multi stage feedback (even if it is only around the input stage and VAS) requires more attention to frequency compensation and use of lag compensation probably. Because we wanted to avoid any source of TIM, we decided to go this way.

Steven

 

[Tube_Dude]

Because we wanted to avoid any source of TIM, we decided to go this way.

Steven

Thanks Steven!

And what about a capacitor across the zenner?..

 

[Steven]

Jorge,

I don't remember whether we tried a capacitor across the zener. Anyway, the impedance of this zener is around 4 Ohm at the chosen bias current. Compared to the 27k//27k load this is very small.
But, you're right, close to amplifier clipping there is significant current modulation in the zener, resulting in impedance modulation, resulting in gain modulation, which is harmonic distortion. Still the VAS distortion is below 0.02% (only 2nd and 3rd harmonics, all other harmonics below the measurement limit of 0.003%) close to clipping. Distortion of the output stage at full load is bigger. But adding a capacitor could be worthwhile. Thanks.

Steven

 

[PMA]

Pavel,

It's not a secret that could be discovered by looking at the image properties. I also mentioned it in writing in the same post. The output stage uses Malcolm J. Hawksford's error correction circuit. Details about this can be found in the JAES, Vol. 29, January/February 1981, p. 27-31. Title: Distortion correction in audio power amplifiers.

I always wondered why so few amplifers use this error correction scheme. I only saw two other designs, one from Robert Cordell, and one from Peter van Willenswaard. It's simple and works very well, although, I admit, it's not very intuitive. As far as I know it is not covered by patents.

Steven

Steven,

I completely agree. I would like to add the name of Pavel Dudek (Upupa Epops here), who has been producing the amps with Hawksford type correction since 1992. Used in a mosfet amp and it brings considerable improvement to the sonic characteristics.

Pavel

 

[Steven]

Hi Pavel,

Do you know whether that circuit is similar to the one from Robert Cordell, who also uses the Hawksford error correction circuit with MOSFETs? Or does it provide other nice features and clever circuit ideas? Has it been published somewhere?
I noticed some remarks about a website of Upupa Epops, but I couldn't find it. That usually means that I did not search long enough, but asking may be quicker.

Steven

 

[PMA]

Hi Steven,

ufortunatelly I do not know the circuit by Robert Cordell, but as I would guess from my often conversation with PD the circuit is inspired only by M. Hawksford. His amps were published in the Czech journal Amaterske Radio in 1992. The amp with cross-over region correction can be found in No.8/1992 and is quite similar to your circuit, first stage uses cascodes instead of CFP's, the output ís with MOSFETs, feedforward correction is also used. I do have the schematics scanned and if Pavel D. allows, I can display it here or on my web page and give a link here.

Pavel

 

[PMA]

Steven, I found Pavel Dudek's amplifiers on Dutch web page!

Here is the link to the schematics:
http://margo.student.utwente.nl/el/ampz/pics/dpa330b.gif

 

[Steven]

Hey, that's funny. Someone keeps these schematics at the Technical University Twente in the Netherlands. The site mentions that the schematics were published in 1992 in one of the biggest Czech electronics magazine. The link to the site www.dpaudio.cz does not work though.

Its remarkable how many similarities exist between his design and ours. Even the VAS stage is similar (with emitter follower, Hugh ) On the other hand his design does have overall feedback. Well, at least our design appeared in 1983 as part of a thesis and the schematics were not published, only part of the background story on distortion in 1984/1985. So it is clear that both designs were developed independently of each other.

Steven

 

[PMA]

Steven,

the designs were developed independently. I know the background of PD's design.

The link to dpaudio does not work as the company does not exist anymore.

I phoned with Pavel D. and he proposed to send the newer circuitry in better image resolution, with PCB as well.

The amp, slightly modified, serves as a reference power amp in our country. Tested and compared with very very well known amps, this is a winner in clarity, purity and resolution. And one of very seldom able to bring enjoyment when listenning to reproduced philharmonic orchestra.

Pavel

 

[PMA]

Its remarkable how many similarities exist between his design and ours. Even the VAS stage is similar (with emitter follower, Hugh )

Steven

It is absolutely neccessary in this circuit and you know why .

 

[andy_c]

(...)I always wondered why so few amplifers use this error correction scheme. I only saw two other designs, one from Robert Cordell, and one from Peter van Willenswaard.

Hi Steven,

The Halcro also uses the error correction technique. I tried looking at the schematics at http://patft.uspto.gov (which I was able to do a few months back) and seemed to be having browser or site problems.

I was playing around in simulation with the error correction technique for a while. Another user here asked me some questions about it in email and the text below was my response

BTW, if anyone is interested in the Hawksford or Cordell papers, feel free to shoot me an email.

"I'm looking to have an amp that requires no output inductor and am trying to design it so it will produce a full power square wave at 20 kHz with no ringing or overshoot into a 2 uF capacitive load, yet still have low sine wave distortion. The square wave requirement into a reactive load has meant having a low unity loop gain frequency, only a little over 100 kHz, so the feedback at 20 kHz is quite low. This is why I was interested in the error correction technique. I have an LTSpice simulation of the Cordell circuit, and am now playing with a simulation of an error-corrected version of an output stage that looks like a mixture of the Cordell circuit and the Holton AV800 (14 output FETs total).

I'm having a tough time with it. If I compensate the error correction loop so it works well with an 8 Ohm load on the output stage, it oscillates with 2 uF. If I compensate the loop so it works well with 2 uF, it oscillates with 8 Ohms. If I overcompensate it so it's stable with both, I get no improvement in 20 kHz harmonic distortion over the uncompensated design - yet there are still odd frequency response quirks introduced.

It appears at first glance that the stabilization of the error correction feedback loop should be easy, since the loop gain is low. But one problem is that the loop in its idealized form tries to force the output to equal the input independent of frequency. So at high frequencies, large error correction voltages will appear that are compensating for frequency response effects. In the ideal case, you'd want to low-pass filter the picked-off input of the output stage with a filter that very closely matches the frequency response of the output stage, prior to subtracting it from the picked-off output to get the error voltage. If the frequency responses matched perfectly, you'd have an error correction voltage that only compensated for distortion and not frequency response effects. But the Cordell/Hawksford circuit cross-couples the picked-off input to both sides of the difference circuit, making this impractical. However, if you look at the schematic in the Halcro patent, you'll see he modifies it so that a level-shifted version of the output is only connected to the base of the error-correction transistor, and the level-shifted input is only connected to the emitter. This allows separate filtering of the picked-off input to match the output stage frequency response. This works great with an 8 Ohm load. But when 2 uF is put in parallel with the 8 Ohms, the bandwidth of the output stage drops dramatically. Now there's a big frequency response mismatch. When this happens, the loop gain of the error correction loop (as simulated using the Middlebrook technique), though it is only about -40 to -50 dB at low frequencies, jumps to about +15 dB in the low MHz range, with unpredictable, uncontrollable phase. This is what causes the instability I'm fighting.

Actually, I'm about ready to give up on it. Even when the compensation works well, it seems to introduce complex poles into the frequency response of the output stage, making the phase drop off more rapidly than one might guess by looking at the magnitude plots. Also, the reduction of distortion with the 7 p-channel, 7 n-channel output stage is not as dramatic as with a single device. This is due to the distortion being fairly low to begin with, due to the higher effective gm from having so many devices in parallel. Also, upping the bias current to 150 mA per device keeps the distortion in check. I can get good transient response without the error-correction stuff, even with a 2 uF load.