Discrete sounds better than integrated? a possible reason

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

ruffrecords

Well-known member
Joined
Nov 10, 2006
Messages
16,155
Location
Norfolk - UK
Following the passing of Rupert Neve I was pondering the longevity of his designs and others of the period versus the myriad integrated op amps now available that wax and wane every few years.. Is it more than just nostalgia that causes people to think those old designs sound better or is there something tangible that actually creates a well liked sonic signature?

If there were a difference, it would have to be more than just the supposed mojo of vintage components; there would have to be some common thread in the designs that sets them apart from your typical integrated circuit op amp. In the past I have postulated that it may be that the most favoured vintage discrete designs are class A. Others have suggested that the transformers are a major contributor (except Helios mixers had very few transformers yet were considered very musical).

It then occurred to me that there may be another factor that makes a significant difference to the sonic signature. Let me explain.

In the dim and distant past, tubes were the only active devices available. Practical problems due to the large dc voltage difference between inputs and outputs meant that the NFB loop could not easily be closed at DC leading to low frequency stability problems. Miller effect at the high frequency end led to similar NFB stability issues at this end of the audio spectrum. These two factors alone meant it was not really possible to make something like a mic pre with a gain that could be set to a wide range of values (the REDD 47 is a good example of this) because the stability criterion at each gain setting is different.

However, the ancients did have a solutions to these problems. They realised the problem was in the fixed open loop gain. To vary the gain you only had to alter the  loop parameters which meant the stability parameters changed with gain. So they cleverly arranged things so that both the open loop gain AND the loop parameters were varied together such that the stability criteria were near identical at all gains. The first instance of this design I can find is in the German V41 which appeared in 1949. The almost identical circuit was repeated in the well known V76 which appeared in 1959.

i do not know if it was deliberate or not, but this technique of varying open loop gain at the same time as closed loop gain was used in all Rupert Neve's three transistor class A designs of the late 1960s and in the Helios and Cadac three transistor and Calrec four transistor designs of the same era.

So, what is it about this technique that affects its sonic signature. In all designs, the open loop response falls off at some high frequency. Which frequency this occurs at depends an many factors but for stability, the frequency at which this begins should be such that by the time the open loop gain as reached unity, the phase shift is less than 180 degrees - this is one of the basic stability criteria that stops your mic pre becoming an oscillator. The bottom line is that with this type of design, this frequency is usually well above the audio band (Rupert once suggested it should be around 75KHz). What this means for sonic signature is the amount of NFB is the same at all frequencies in the audio band and so the distortion at any frequency is the same. Hence any distortion that is introduced does not alter the relative levels of harmonics in the original sound.

Moving on now to integrated op amps we find they are built differently. First they have enormous open loop gains (100dB is not unusual). They do not have the ability to alter the open loop gain as the closed loop gain is altered so they have a serious high frequency stability issue. To fix this and make it easy for almost anyone to deign with them, they include a capacitor inside the chip that starts the HF roll of well before the top end of the audio spectrum - for instance the TL072 starts it at about 12Hz. This allows you to set the closed loop gain easily with a couple of resistors without having to worry about it oscillating. But what is the price? It is that the open loop gain drops at 20dB per decade over nearly all the audio spectrum. This means that the amount of NFB falls at the same rate which in turn means the distortion rises at 20dB per decade. Hence the distortion introduced does affect he relative levels of harmonics in the original sound. In particular it emphasises the higher harmonics and as we all know, the higher the harmonic the less the ear likes it.

So it is my suggestion that this fundamental difference in design may well be what separates vintage discrete designs from modern integrated circuit ones. As a caveat, I will add that not all op amps are as bad as the TL072 in this respect. For instance, the internal capacitors of the NE5532 start the HF drop at 1KHz.nearly 100 times higher than the TL072. Perhaps this goes some way towards explaining why the NE5532 is considered a relatively nice sounding op amp.

All the above is simply an observation based on what I know of the various designs themselves. I am not saying it is certainly an explanation for the sonic difference between the two types of amplifier, only that it is a possible explanation.

Cheers

Ian
 
I'm actually a fan of a fairly fixed gain.

But on to your theory: I know you know this but, take the 1073 - The gain of the input stage is actually not varied much, some 10dB.  The settings below lowest gain are merely attenuation pre the amplifier proper.
On high gain settings, they switch in another gain block. 

And the output amp keeps a steady gain of some 16dB + another 4dB from the output transformer.

Or are you saying that the open loop gain was first modified with degeneration/compensation etc.  so as to provide a specific (lower) amount of negative feedback? 
 
P.S.  - the way that the feedback is changed in the 1073 input amp will tend to show the same frequency response characteristics on all 3 gain changes.  This because of the way feedback is applied.
These days it's called current feedback, although it isn't technically current feedback.  That just seems to be the now adopted way of defining it.  I'm only going along with naming convention despite knowing differently. 


Anyway...    ;)
 
Winston O'Boogie said:
My belief:  the main difference between things is usually down to how they misbehave.
I think that is in essence what I am saying. Integrated op amps misbehave in an atonal way. And the reason is they way they arrange the combination of open loop and closed loop gain.

Cheers

Ian
 
Thanks for that Ian ,

Just to add a general observation to this , Ive typically found the more feedback the more linear things are as long as you dont approach overload , high feedback and the onset of clipping turns bad very quickly , the amp is trying to correct but cant , 'latching' was the term associated with big ugly poweramps , but im sure the principle holds true with all high 'feedback' amps. A profligacy of gain which needs feedback to keep it under control doesnt offer the most gracious overload characteristic. I guess the effect depends to some extent on source , an electric guitar miked up from a tube amp is going to have all those transients gobbled up and nicely limited so it makes it easy to find the right gain setting , a vocal on the other hand might have short term peaks 10-15 or 20 db above nominal , a good tube mic will help here where something like a U87 might benefit from compression more . In the good old days of valve gear specs of upto 5% thd were common , mainly second harmonic of course , with modern high feedback topologies and op amps 5% thd would be a very painful experience .

And thanks once again for drawing attention to what is most a neglected aspect of modern audio design .

ahahah Winst , thats it , you want your gear to misbehave gracefully not disgracefully  ;D
 
We like the way some of the stuff with transformers, tubes, 3 transistor gain stages with simple resistive loads, you name it... Misbehaves.   

Keep an op-amp from misbehaving and, ideally, it should sound like nothing. 
Cookie cutter audio.  No art in it.  Fine for lots of things.

But some stuff can be as much art (or accident!) as science.  Whatever gets you and, hopefully others, off! 

Man  8)
 
I'm not sure I follow. Regardless of open loop gain and feedback levels at different frequencies, why would we not be able to see that in a simple distortion measurement?

There are many reasons for differences in how something sounds but I think it mostly boils down to psychoacoustics. If the gear looks cool it will sound good. Even if bandwidth is limited and distortion is relatively high it can still sound good or even be perceived as better (consider The Animals and Bo Diddly and many other late 50's / early 60's recordings that still sound great).

However, I do think that it is possible that there could be some transient / modulation phenomenon occurring that could make a recording sound different and be perceived as bad or good in some way. This is not something that would be measurable looking at the FFT of some constant amplitude stimulus. I think there are things going on at low frequencies that are overlooked.
 
Indeed, current FB opamps have some nice inherent advantages, particularly very high BW that results in low THD and almost no slew rate limitation, together with almost zero stability issues as long as the NFB path is free of capacitive effects.
An interesting comparison would be between a discrete and an integrated version of the same circuit.
Indeed, the fact that the distortion reduction is constant over a large gain range is commendable, however the corollary effect is that a decent voltage feedback amp has a more limited gain range where it excels, and that's probably why some are so appreciated.
I would think the almost-cult following of the API312 on drums is partly due to the fact that its structure is optimized for medium gain.
 
;D you troll, you  ;D

But thanks, Ian - some very interesting ideas in there. It's clear to me that there certainly are subjective benefits to some of the old technologies, but until now a complete lack of believable explanations.

/Jakob E.
 
wasnt everything class a back then?
i think in combination with the fact that there were  VU meters everywhere instead of led meters, forcing the operator to maintain a sane gainstructure that was focusing on rms content instead of peak content, was responsible for alot of good old sound. Today i see and hear,  all the time, people slamming everything towards red on there peak meters with terrible results.
 
also, i think Rupert used transformers using nickel (atleast input) which gives lower distortion at lower levels if i am right.
 
Winston O'Boogie said:
My belief:  the main difference between things is usually down to how they misbehave.

Modulations, swirls, clipping and asymmetry. I always thought that it came down to misbehaving nicely (pardon the oxymoron).
 
5v333 said:
wasnt everything class a back then?
Not necessarily. But yes, very often. It was almost mandatory, particularly for circuits that did not use NFB.

i think in combination with the fact that there were  VU meters everywhere
Could it be that the distortion induced by VU meters was part of the sound signature?  :)
 
abbey road d enfer said:
Not necessarily. But yes, very often. It was almost mandatory, particularly for circuits that did not use NFB.
Could it be that the distortion induced by VU meters was part of the sound signature?  :)

hmm.. do you mean that the VUs interfered from bad isolation/buffering?
i havnt thought about that...
 
5v333 said:
also, i think Rupert used transformers using nickel (atleast input) which gives lower distortion at lower levels if i am right.
that's pretty much the case of any decent design.
High Ni content is a requirement for a good input xfmr.
It's only recently that winders offered cheap input M6-core xfmrs, which have a different sonic signature.
 
well in that case i guess it must have been a factor if Rupert didnt isolate his VUs. did he?

sorry  if i was dragging the thread away from the main point that Ian was making about feedback/gain setups.
 

Latest posts

Back
Top