ODNF can be looked at as an ordinary global N Fdbk setup where the input signal has been moved up in the amplifier string closer to the output. Leaving the front end high gain stages with only the N Fdbk signal to handle. The error signal still re-circulates thru everything like in a global form, so spectral spreading is still occurring. Now one has to adjust a trimmer to get the N Fdbk matched to the input "effective" level, and this depends on gain B staying constant and without distortion (zero!, 0.001% or less). The availability of National LME type Op Amps is what makes this all practical.
While the theorem that global N Fbdk is more effective per gain available is true, it is limited in a tube amp by the OT freq. response. The OT magnetizing current, tube dist., and winding R (and crossover distortion) are the main culprits to overcome. Not well appreciated however is that local N Fdbks can fix these problems readily (well, fix the primary R well), and with good freq. response.
So we are left with the typically low dist. front end gain stages as the only difference between global and local N Fdbk schemes. Ordinary global N Fdbk has the front stages inside the loop.
ODNF uses an LME type Op Amp to replace them.
Local nested schemes can just use the available OT BW in global N Fdbk to fix the front end and fix the OT secondary R (also the primary R if not already corrected by a local fix). The local N Fdbks fixing the tubes, crossover, and the OT primary (and magnetizing current).
This last approach does not require an expensive OT, and can easily be done with all tubes. Nested Fdbk loops look like the winner to me for a tube amp.
While the theorem that global N Fbdk is more effective per gain available is true, it is limited in a tube amp by the OT freq. response. The OT magnetizing current, tube dist., and winding R (and crossover distortion) are the main culprits to overcome. Not well appreciated however is that local N Fdbks can fix these problems readily (well, fix the primary R well), and with good freq. response.
So we are left with the typically low dist. front end gain stages as the only difference between global and local N Fdbk schemes. Ordinary global N Fdbk has the front stages inside the loop.
ODNF uses an LME type Op Amp to replace them.
Local nested schemes can just use the available OT BW in global N Fdbk to fix the front end and fix the OT secondary R (also the primary R if not already corrected by a local fix). The local N Fdbks fixing the tubes, crossover, and the OT primary (and magnetizing current).
This last approach does not require an expensive OT, and can easily be done with all tubes. Nested Fdbk loops look like the winner to me for a tube amp.
Last edited:
A modest improvement to the ODNF scheme seems feasible. (well, uPID might do even better)
Instead of attenuating the output signal and subtracting the input signal to get the error signal and then amplifying back up. Leave the output signal as is, and Op. amplify the input signal up to it. Then do the subtraction. Now the error signal will be larger, so will need less gain B to lower distortion, or using the same gain B will lower dist. even further. Just an idea.
Once one has a super low distortion Op Amp, like the LME type, one could try the Yamaha ZDR (feed-forward) scheme. Drive the error signal with the Op Amp and sum with the output using a -very- low output Z xfmr (likely ferrite, one turn secondary) in series. The Op Amp only has to supply distortion correction power. Very low if the main Amp is already low N Fdbk corrected.
But then, well, all these high correction schemes are going to sound like a super low distortion SS amp. Is that the objective?
I'm really more in favor of maximum cleanliness with minimum parts. Or maximum effective simplicity.
Instead of attenuating the output signal and subtracting the input signal to get the error signal and then amplifying back up. Leave the output signal as is, and Op. amplify the input signal up to it. Then do the subtraction. Now the error signal will be larger, so will need less gain B to lower distortion, or using the same gain B will lower dist. even further. Just an idea.
Once one has a super low distortion Op Amp, like the LME type, one could try the Yamaha ZDR (feed-forward) scheme. Drive the error signal with the Op Amp and sum with the output using a -very- low output Z xfmr (likely ferrite, one turn secondary) in series. The Op Amp only has to supply distortion correction power. Very low if the main Amp is already low N Fdbk corrected.
But then, well, all these high correction schemes are going to sound like a super low distortion SS amp. Is that the objective?
I'm really more in favor of maximum cleanliness with minimum parts. Or maximum effective simplicity.
Last edited:
May I ask you how do to prefer to manage the nested feedback loops? output tube plate to grid, output plat to driver cathode, plus gnfb? Some positive current plus negative voltage feedback plus local feedback (à la Wavebourn)? Something different?
Will this be adaptable to a "paraphase concept" phase splitter, with the voltage divider referenced to the secondary of the OPT instead of ground?
Last edited:
There are various advantages and disadvantages to the different local Fdbk models.
Plate to driver cathode doesn't require a Fdbk criss-cross for phase (P-P), but then it has a large voltage difference so usually requires a cap(s).
Plate to driver plate requires current drive and gives low Fdbk loop gain. It also has a distortion residual with low gm output tubes, due to grid V affecting the N feedback current. UnSet/CED seems preferable for low gm output tubes.
Plate to driver grid makes for low input Z at the driver but good gain for the local loop.
I prefer (in theory anyway) to do plates to driver screen grids (P-P), so the driver grid 1 stays high Z. This has several drawbacks however, crossed Fdbks for phase, and either screen grid Mosfet followers, or scaling the screen grid V variation to a constant fraction of the output tube grid V variation to get constant Z at the screen grids. (driver screen grids at a constant fraction of the driver plate V, intercepting a constant fraction of plate current ) And the driver tube needs to have a very good interal triode mode. This can be restrictive without the right driver tube.
UnSet/CED seems useful for single tube N FDBK. And it can provide nice triode curves, even for low quality triodes (like regulator tubes, say). And low output Z as well.
The output current Fdbk for critical speaker damping seems interesting, but seems not to be favored these days.
Some global N Fdbk seems generally useful, unless the output tubes are low Z triodes (regulator tubes say). Might as well use the OT phase room if needed.
Then there is Crazy Drive, which is actually Feed-forward. It can get near constant gm from pentodes/B. pentodes. It needs some further Fdbk to get the output Z down. And cross-over distortion performance seems unproven. I think it should be able to do as well as grid 1 drive here, but at much lower idle current.
-----------------
I guess, I don't see why not, would have to work up a schematic probably.
Last edited:
smoking-amp,
Thanks for shedding some light on various methods and tradeoffs of negative feedback, and other linearization schemes.
Thanks for shedding some light on various methods and tradeoffs of negative feedback, and other linearization schemes.
Could also add that all schemes for local feedback in an output stage, fedback series or parallel, extract a toll from the driving stage, in driving current and/or driving voltage, so extract their pound of flesh, complying with the Second Law.
An extreme example, like a cathode follower final, requires a driver that's more difficult than the output. A half-extreme example, like a McIntosh, circlotron, Crowhurst, is almost possible, with a lot of fancy footwork. An only slightly extreme example like the old Quads could be done pretty conventionally.
Parallel feedback designs are post-modern. They usually have the important part right, voltage sensing feedback taken from the point of drive to the output transformers. Applying the feedback in parallel to drive current ("Schade") is easy to understand, easy to modify on the fly, and can usually work really fine. But it's not optimum for distortion, because it loads the driving stage.
Two stage amplifiers have potentially the lowest distortion and highest input impedance (only matters following a phono cartridge or a volume control) with feedback brought back to the first stage's cathode. Getting the DC voltages right means some cypherin' and some gozinta's but is doable.
All good fortune,
Chris
An extreme example, like a cathode follower final, requires a driver that's more difficult than the output. A half-extreme example, like a McIntosh, circlotron, Crowhurst, is almost possible, with a lot of fancy footwork. An only slightly extreme example like the old Quads could be done pretty conventionally.
Parallel feedback designs are post-modern. They usually have the important part right, voltage sensing feedback taken from the point of drive to the output transformers. Applying the feedback in parallel to drive current ("Schade") is easy to understand, easy to modify on the fly, and can usually work really fine. But it's not optimum for distortion, because it loads the driving stage.
Two stage amplifiers have potentially the lowest distortion and highest input impedance (only matters following a phono cartridge or a volume control) with feedback brought back to the first stage's cathode. Getting the DC voltages right means some cypherin' and some gozinta's but is doable.
All good fortune,
Chris
I've simulated results of the "Corr Diff" and are very promising:
Shunt Cascode Driver meets UNSET for Push-Pull
I add the link here to be able to follow the results on that circuit and develop that circuit even further for anyone interested (you can find the PSpice file attached to that post).
Shunt Cascode Driver meets UNSET for Push-Pull
I add the link here to be able to follow the results on that circuit and develop that circuit even further for anyone interested (you can find the PSpice file attached to that post).