Neve-ish modded GoldenAge-pre73 w.Carnhill inp TX damping results/advice?

[HenryL]

Hello people,

Wondering if you can help me with interpreting some mic-input TX damping efforts...

I decided to tinker with my Golden Age Pre-73 (mk1), replacing the mic input and the output transformers with Carnhills (as well as changing some caps etc.) I've looked at various posts and didn't get a convincing picture on the zobel/input damping components with the Carnhill VTB9045 tx in the GA circuit. Some people just replace the original GA TX with the Carnhill without changing anything else (leaving the original GA parallel 12k res and 1nF cap in place evidently). I decided to initially try it as per the Neve 1073 schematic with just 180p across the TX output. This is also as in eg. Madrian's 1073ish clone using the same Carnhill TX.

Then I thought I really should take a look at it with an oscilloscope...

This is what it looked like with 180p across the Carnhill (a la Neve / Madrian etc.) :



(This was with square wave input signal ~6kHz, 0.1V, Zout ~150ohms following the approach described by Dale Roche of Jensen as posted elsewhere around here somewhere.)

I removed the 180p and tried first of all just adding a resistor across the TXout. Again from the Dale Roche advice this might typically be about 6K (1500R times the impedance ratio). I tried 22K then 10K and the 10K already seemed to be doing more than enough so I stopped there:



Then trying various caps in parallel with the resistor I settled on 250p, which tightened the corners up a little:



Now (above) it has got a little visible overshoot and ringing back again, but it varies depending on the gain setting, for example on another gain setting it looks like this:



So... this looks plausibly OK to me but as I don't have experience of this maybe it's bad ! Can anyone throw any wisdom and experience on this result?

I've not tried it in action yet as the rack it's normally plumbed into is still in pieces, so it's all theoretical at this point.

I was surprised how much difference there was between some gain settings. I didn't confirm the GA gain selector design was exactly the same as the original Neve circuit -as checking it out would be a pain- but since the rest of the circuit is essentially the same I'd be surprised if it's different. (?)

Then I found this at JLM audio: Neve transformer info

In the above JLM info I found it a bit hard to tell which trace was the Neve, but it seems they are all a lot more ringy than I achieved above, so have I in fact over-damped it in spite of the advice to try and get rid of the ringing?

 

[HenryL]

ps. looking at the above JLM graphs today when my computer screen was not on 'night setting' helped
I also realise now that I was looking for the wrong Neve transformer plot and the 10468 is the yellow trace in the middle of the bunch. So my own results may not be so very far off target. I'm still wondering if I have it a little over-damped on some gain/sensitivity settings or whether this is in fact an exemplary-looking trace!? (Awaiting registration on the JLM site to seek some additional info there).
Anyone?

 

[Matt Syson]

The fine tuning of transformer loading was usually done on test and certainly was when I was doing this sort of thing in the 1970's. Compensating for a transformer response has to be done when viewed as a complete system not just taken in isolation although having a table of 'usual suspect' values perhaps with series resistance and a capacitance is a good start. For output transformers you have to define the load it will drive (say 600 Ohms resistive but also the typical capacitance of 50 feet of cable (for a small studio) or 500 feet in a broadcast setup.

 

[HenryL]

The fine tuning of transformer loading was usually done on test and certainly was when I was doing this sort of thing in the 1970's. Compensating for a transformer response has to be done when viewed as a complete system not just taken in isolation although having a table of 'usual suspect' values perhaps with series resistance and a capacitance is a good start. For output transformers you have to define the load it will drive (say 600 Ohms resistive but also the typical capacitance of 50 feet of cable (for a small studio) or 500 feet in a broadcast setup.

I take your point about it being the complete system input => output that we are interested in ultimately. It seems reasonable though when it comes to ringing/overshoot in an identifiable part of the circuit where this is known to occur, to work on this stage ? The part of the circuit being examined here is not isolated but is driving the load of the following transistor gain stage. I probably should have made that clearer. (edit: the same goes for the info on the JLM site which is also from in-circuit probing not in isolation)

I did arrive at values through a testing and component substitution process, which was when I realised I was no longer so sure what 'the right result' looked like ! (hence my question )

Cheers / Henry

 

[HenryL]

Forum platform malfunctioning ? I got the following response from aurt by email but it does not actually appear here on the forum ?! Anyway, here is the content of aurts post from the email:

aurt said:
"The resistance across the secondary terminals of the mic transformer in the 1073 varies over the gain settings. There are resistors in the gain switch and the input amp, and there's the input impedance of the amp. This likely explains your differing results. The 10k in parallel with the other resistance would make for different loads. Without the 10k, 50dB and up should be fairly consistent, at somewhere around 28k, ignoring some local feedback that might change that some. Below 50dB, it varies quite a bit, down to around 4.8k at the -20 setting."

Dohh - of course I should have checked the actual attenuator design. I slackly assumed that being a multi-pole switched design that it would be some kind of pad maintaining a consistent Zin towards the tx - I'll check the schematic, but I feel sure you're going to be right about that!

"I would think it would be more pronounced with just the 180p cap in there. I'd be curious to know. I'm guessing that the 3rd pic is a higher gain setting than the 4th. Is it, or the other way around?"

The first pic is just with the 180p in there and yes it is more ringy. Because of the way I did things initially I didn't pay much attention to the actual attenuator setting at this point - I just set it to get an output LED meter indication of approx 0dBu output. IIRC this was around the -30 to -35 sensitivity position.

3rd pic vs 4th pic - sorry I didn't make a note as to which gain setting was which for these pics - but the reason for not noting it was that I found similar variations when going both up and down in sensitivity from the position I originally did the test in, in other words it wasn't consistent that going one way increased the ringing and the other way reduced it. In general the -55 to -80 positions using the extra gain stage were more consistent but I was slightly dubious about interpreting those as the transistor circuit would have been overloaded.

"I suppose the 180p was chosen as a best compromise for all gain settings, where they all fell into an acceptable range."

I think this must be right. But as the original Neve is not based on exactly the same transformer as I'm using then if trying
to match the general Neve behaviour as closely as possible there is still a question as to what gain setting to base it around
and what sort of overshoot/ringing at whatever setting that is...

"Also wondering if the 10k resistor alters the gain at the output much. Probably also varies with the setting."

With just the 180p across the TX output then the 0.1V test signal was about 0.185V at the TX output, with the setting setting around "-30" to "-35".
With just the additional 10k load resistor across the TX output, no cap, this load pulled it down to about 0.15 so nearly 2dB loss at the input stage compared to just the 180p.

"Also, if you're going to compare to the JLM data, you might as well try 500Hz as well and see how that looks. Again, I'd be curious to see."

I wondered about that, but as essentially looking at the step response of a single step I didn't see why the frequency should make much difference,
within reason perhaps.. (?) I'm also not testing at the same test signal level as in the JLM data with JLM at .4V, mine at .1V
(mine is fixed at that voltage). That suggests the JLM data was probably taken on a different gain setting to my tests.
I've asked on the JLM forum what gain setting was used but not sure if this will get an answer as not really related to any particular JLM product.

Cheers

EDIT: PS I don't yet have personal messaging rights so perhaps someone who does could let aurt know that their response did not vanish completely Also, I will rerun the tests with more careful noting of the sensitivity settings etc. and post them here in due course.

 

[HenryL]

The Golden Age Pre-73 (Mk1) and Neve 1073 sensitivity attenuators are NOT the same, and they produce slightly different variations in the input impedance seen by the Mic input TX at different settings:



A3 = Neve BA283 board, A4 = BA284, or the equivalent stages in the GA Pre-73. "TXout-->Zin" is the resulting input impedance seen by the TX looking into the sensitivity switch.
ZinA is the input impedance of the BA283, BA284 or GA equivalent amp circuit. Ref Neve 1073 input impedance
The GA has different gain staging with slightly higher values of gain adjusting resistors A3adj so the ZinA for the GA at -50 and -45, and the resultant overall Zin, are a guesstrapolation and not to be leaned on too heavily (greyed in the table).
The measurement of the 'Tx in' and 'to Gnd' resistances in the GA wasn't super accurate, but shouldn't be too far out.

The general observations at this point are that the Neve keeps the Zin more consistently close to 4k8 than the GA does, but they are not wildly different from each other. However the GA does have much bigger differences between -30 / -35 / -40 which is in the area in which my TX damping tests were mainly hovering, and would seem to explain the surprising variations found between adjacent switch positions.

Another conclusion would seem to be that since the Neve holds the Zin more constant then it is less important to know exactly which gain setting the JLM traces were taken at - the traces should be more consistent and could probably be taken as a general target, however there is a big difference between the two ranges (-80 to -55) and (-50 to -20) so we'd at least want to be sure we knew which range the JLM info is based on. Though it is almost certainly the (-50 to -20) range given the 0.4V test signal used.

 

[HenryL]

Hope you're all enjoying this. After another series of oscilloscope tests - powered off, powered on, different gain settings, with and without mic tx cap (I used 172p this time it being fairly close to the 180p of the Neve circuit) but no additional parallel load resistor, with and without 600ohm output load, looking at the signal at the input TX output and the final line output, and a few others - my conclusion is that the best overall behaviour seems to be achieved with NO additional components across the input TX, just straight into the sensitivity selector, with this transformer.

The worst selector positions for overshoot/ringing are consistently - in the GA at least - the -40 and -45 positions, and below (below meaning -50 to -55 and beyond), though it improves slightly again below the -45 position. The previously posted effective Zin figures (based on someone's spice modelling) do not really give any obvious clue as to why it would be those positions in particular. I guess it's a more complex interaction with the input of the amplifier circuit. Below -40 the top leg of the potential divider drops suddenly from 2K to 665 ohms (even more in the actual Neve) so the resistance between the TX and the amplifier input is much reduced.

Monitoring at the line output (after the final output-TX) then it's very clear that a 600ohm load soaks up a multitude of sins (also possibly some enjoyable racket) at the output stage, removing all but the most stubborn overshoot spikes in my comparisons.

I took a load of quick photos of the scope traces for most of my test permutations, if anyone's interested in seeing some of them then ask.
When I get a moment I may post some selected highlights.

 

[HenryL]

This time looking at the LINE input I found the same kind of thing. The behaviour seemed better without the neve 2n2 cap across the line input TX. There is still a little overshoot/ring but it's much less.

With the 2n2 cap a la neve:



Ignore the '5microseconds' that actually applies to the next pic, with a triggered expansion at 5 microsecs per div.:



I took the output here from the top of the output level trimpot as the additional gain gives a better picture, it looks the same if taken from the tx itself, just smaller...

Then finally here is how it looks without the 2n2 cap:



There's still a little overshoot and ring but it's much less.

So I fail to see the advantage of having these caps.

A caveat here is that I did not replace the line input tx with one known to be close to the original Neve, but my reasoning is that since the line input circuitry in the Golden Age version is actually identical to the Neve (the attenuator resistor values in the sensitivity selector, the 2k2 additional parallel load resistor and 2n2 cap) then they evidently felt their line tx was reasonably similar to the neve (whereas they selected different values with their mic tx).

So I'm back to the question as to why did Neve put these 180p and 2n2 caps across the input TX's. I've removed them and am going to plumb the unit back in and see if I have somehow noticeably ruined/improved the sound...