Global Negative Feedback on the PP18

29/12/13        Global Negative Feedback on the PP18
I returned to the noisy rectifier hash ridden PP18 problem as it was still bothering me immensely not knowing the exact cause, and not having a suitable fix more so.
I re-read the global negative feedback pages 164 and 188 of Blencowe but was still wondering if it was possible to use the same method of attaching the bottom of the output stage cathodes to the OT secondary’s as I did for the Aston single ended amp:

Above – the NFB method for a Single Ended amp without having to calculate a feedback resistor value – just get the phase correct – swap leads at easiest place (i.e. brown/blue or grey/black if howling occurs (switch off immediately!).

I was curious to know if this could be done on a Push Pull amp as Blencowe has not mentioned it, and after loads of headaches trying and failing to visualize how it would work (in terms of electrons and relative voltage levels), I decided the worst case would be howling feedback and possible damaged valves, so I decided to try it anyway, using the LTP stage so that any hash signals and noise in this stage would also be cancelled, hopefully, and save me finding suitable resistors for the voltage divider method explained later – I tried it but got howling feedback as I had not checked the phase relationship between the OT secondary and the LTP grid pin2.
I reconnected for normal use then swapped the wires at the power tube anodes then I went through the PP18 circuit from end to end like I did for the Maggie to be sure and marked phase relationships relative to the input and also confirmed for myself there is indeed NO signal at the pin 7 grid of the LTP V2:

I reconnected the 16 ohm output jack to attach to the bottom of C6/R6 after de-soldering them from ground again. This gave an out of phase signal from the OT sec to the previously static DC point at pin7 of V2 via C6 – the same phase that is at pin2. This worked for the lower power 190V mains PT secondary to give what appears to be HEAVY NFB to the circuit as it was almost completely hum and noise free at full volume with no test signal. The downside is there is no LTP distortion tone available with increased volume because the NFB has squashed the gain so much and prevented the saturation of the LTP stage.
When I switched to the 275V higher voltage PT secondary, I got the phenomena that I think is termed “motor-boating” – caused I think, by the phase drifting of lower frequencies so they cause PFB instead, from what I have read in Briggs. This shown in the video below:

So, the NFB worked from a noise cancellation viewpoint – but a little too well overall! The NFB is really heavy because, I think, of the relatively high voltages from the OT secondary compared to those in the LTP stage giving almost maximum possible signal cancellation.
I tested again by attaching to the lower voltage 8 ohm tap to see if the NFB would be lessened and the motor-boating would still happen at 275V which it still did. More research required on this…
I tried this method on the power stage also afterwards, but it didn’t remove hash noise so is pointless in this amps case, other than to know that it is possible to use these NFB methods. The LTP connection proved a very interesting exercise for the 2 main reasons above of very flat FR response and discovering motor-boating, so though I haven’t solved my main hash noise problem to my satisfaction – by removing noise but retaining some LTP stage overdrive, I have learned much by experimenting.
On the PP18 schematic this NFB method requires connecting the LTP tail to the 16 or 8 Ohm taps of the OT secondary as below:

Above – the same principle as for the SE amp, disconnecting V2 tail from ground and connecting to 8 or 16 ohm taps.

FR graphs for 16 and 8 ohm taps

I think this is the flattest FR plot you will ever see on a homemade valve guitar amp! – more like HiFi:

Compare that to an FR graph when the PP18 was first built:

To explain the usual logic and method behind NFB applied to an LTP stage – thanks to Blencowe as usual – is that a voltage that is in the SAME phase as that at the signal grid (pin 2 of V2) is applied to the opposite GRID (pin7 – a fixed DC point with no AC present normally), via a suitably calculated voltage divider chain, so it opposes the main signal as each side of an LTP conducts in opposition e.g.:

Unlike the SE version above where a same phase NFB signal is applied to the cathode, the NFB for an LTP is applied to the grid via the capacitor Cg2, either after doing some horrendous calculations using actual circuit component values, such as those on page 188 of Blencowe, or from actual voltage measurement. I find it easier to understand by using actual measured voltage ratios and see them in proportional terms adding or subtracting from the pre NFB signal voltage.
When considering this option, I set a 200mV signal at the LTP gate using the volume control, then measured a corresponding 6V signal at the 16 ohm tap of the OT secondary. My logic for using the voltage divider method was that as this is a gain ratio of 6V/0.2V = 30, I would try a divider chain that would feedback say a 50% reduction, so halve the normal pin2 grid input to 0.1V, a 60:1 chain of say, a 600K and 10k resistor.
This worked as NFB and gave a 1/3 reduction in voltage at the OT secondary from 6V to 4V. Unfortunately it did not remove the hash and hum which is the whole point of all this.
A 50% signal reduction at the grid in dB terms is 20 x log0.5 = -6dB so is the lower end of Blencowe’s recommended 6-12dB max NFB values before instability and oscillation may arise again.
I tried again with the highest 12db NFB figure reduction at pin2 using this standard method and the following logic:
In Briggs table below, 12db is a 4:1 ratio so a voltage divider would be needed to reduced the grid from 0.2V to 0.05V, so a voltage of 0.2-0.05V = 0.15V would need to be applied to the opposite grid right? To reduce the 6V at the OT tap to 0.15V, a 6/0.15V or 40:1 divider could be tried e.g. a 39K and 1K resistor chain.
After trying this and still having hash noise, I resorted to trial and error with the following values:
300k:10k divider at 16 ohm tap – Hash noise with little higher volume distortion
168k:10k divider at 16 0hm tap – Hash noise with little higher volume distortion
80k:10k divider at 16 0hm tap – No Hash noise, but no higher volume distortion (no good either as I don’t want a totally clean amp)
56k:10k divider at 16 0hm tap – No Hash noise, but no higher volume distortion
100k:10k divider at 16 0hm tap – Hash noise with little higher volume distortion
132k:10k divider at 16 0hm tap – Hash noise with little higher volume distortion – bearable compromise between hash and some distortion.
At this point I decided to check direct noise and volume level comparison against the Maggie to see if I was getting anywhere.

Summary reminder of why NFB is (generally) desirable:

If I ever doubted by how much #2 is possible, my FR response plots above certainly convinced me how much linearity can be increased! I just need the tweak to stop the low frequency motor boating now.

Checking against actual circuits like the JMP Marshall and the Carlsbro, it seems ratios of 10:1 and 20:1 are used:

Above – a 1k to 100R divider ratio of 10:1 in the Carlsbro 60.

Above – a 47k or 100k to 5k presence pot ratio in the JMP – either a 20:1 or 10:1 divider.

1/1/14 – After sleeping on it all…
Ok, by now you probably know how stupid I can be – as I realised myself this morning – why not just put in a variable pot – which I have actually read suggested in Blencowe, instead of trying different resistors – and keep it as a feature? That way I can have a variable feedback amp that goes from no NFB, to give the normal hash noisy, higher gain amp with its designed distortion, to heavy NFB giving a super flat FR, super clean and quiet only, low gain amp. Duh..!
Pondering this option I realised that yesterday I had made a little circuit error when connecting the resistors as I had not reconnected the 56k LTP tail resistor to ground from when I had used the full feedback method, so technically had messed up the LTP biasing as I was adding it on top of the chain instead of having just the cap to the chain, whilst testing the NFB divider.
It didn’t seem to make much difference really then, but made me think harder about how to connect the variable pot correctly so that I would always have a minimum resistance to ground for when the pot was at minimum to limit shorting out the OT secondary to ground. I went with the 10k 10W resistor I had used to experiment as this would also ensure minimal current would split off from being in parallel with the speaker.
I added a 0.5M pot and I now have the circuit below:

Note the 0.1uF value shunt cap with the 5K pot in the JMP circuit example further above is for presence frequency – to cut or boost NFB for mid frequency values only, not for global NFB reasons as I am looking for here. The principle is the same though.
This seemed to work OK, and removed the motor-boating on the 275V setting also, so I checked the overdrive, general noise levels and tone, then against the Maggie:
PP18 NFB Summary:

Maggie OD Comparison:

All I want to do now is get some voltage numbers at the 16 ohm tap for max power output levels to see the power difference between these amps, and the PP18 at max and min NFB.