|
< <
BACK TO DIY MAIN PAGE
I
originally promised two suggested versions, one with Tubes and
one without. But as of November 2004, the non-buffered version(s) have been withdrawn.
Actually the NON-tubed
version had many more complications though it looked like a simpler circuit. But the idea of this project/website was to promote buffered IGCs.
Let’s look at Thorsten’s original IGC:
Now this is not meant to be a critique of Thorsten’s
circuit perse’. Rather it is a logical starting point as it was originally
for me. About a year ago I was approached by a friend with limited DIY
abilities (hope he doesn’t mind me saying this) to build the above circuit.
Which I did and found it surprisingly good. But I also started to examine it
in more detail and was convinced that there were areas that could be looked
at. For example, notice that Thorsten chose a LINEAR volume pot, not the
usual Log. This was actually quite clever: The pot is actually in the
feedback path and results in the pot behaving more like Log. But it does this
by shifting feedback and gain.
Here are the components/elements that control Volume
and Feedback vs. Gain, all are inter-related.
Let’s assume that the source is Lo Z, such as a CD
Player, typically 200 Ohm, or small low fraction of the 100K pot. Let’s
assume Zero Z for simplicity. If we were to have the wiper at the top (max)
then the input impedance is 100K and 10K in parallel = 9K09.
This circuit is –1.5dB at 10Hz, I would prefer it
being flat to 10Hz and no more than -.5dB at 5Hz. Why? Because this is a
feedback amp and I want to minimize the potential added phase shift that
this causes inside the loop. It has been my past experience that this
is an audible improvement and also can better the sense of timing as well.
Possibly these values will cause even MORE noticeable
LF roll-off with AC coupled sources, such as tube line output stages. So I
would prefer the 2u2 cap be quite a lot higher value. I would like to see 10uF. Response flat down to 10Hz
would be my aim and by changing other values, only 3u3 will be needed.
Let’s look at the way feedback varies with pot
changes. The gain is set by values 220K/10K = 22, or 26dB, – but
this potential max gain (hence lowest feedback) is reduced by the pot
changes. At pot max position the 100K pot is not seen because of ‘Zero Z’
source and no change in gain/feedback. Same applies when the pot is set to
minimum (grounded). But what if pot is in the mid position, as this is where
it is more to be when in general operation?
The basic maths shows 100K pot will become 25K
(because it is ‘Zero Z’ at ground too so it’s 50K in parallel with 50K =
25K) in series with 10K, hence gain 220/(10+25) = 6.28 or 16dB.
So feedback changes and in the mid-position we have
10dB more feedback. (NS also specs minimum closed loop gain of 20dB for good stability).
Now in basic op-amp practice that ain’t supposed to
matter. That’s how you set gain, so it’s perfectly legitimate to vary
feedback – and with a linear pot it works to advantage, as the pot now seems
to operate similarly to a Log pot. I feel sure this guided Thorsten’s mind?
If not he may set me straight.
But it also means you cannot set the feedback exactly
where you want it and you can’t easily reduce it. I note that the original
47 Labs Gaincard has fixed gain slightly above 30dB, so the feedback values
they must have used were not the ones shown in National Semiconductors PDF
data file (20dB). I believe they did this because it sounds better. But in Thorsten’s circuit the gain drops a whopping 10dB which means that typically
14dB + more feedback relative to the Gaincard.
Using a buffer is the ideal here and hence no pot in
the feedback path. For further discussion, I suggest you read my essay
“Tubes & The Gainclones”
Using a Unity Gain Buffer (i.e. no gain or gain = 1)
looks like this:
Now there’s no variable feedback. Gain is now a stable
22 times or 26dB. But feedback value resistors should be adjusted to give
gain a bit above +30dB a la Gaincard and predictable LF and HF roll-offs
implemented. What could be better than using a tube here? It might well
benefit the overall sound anyway (it does!). The actual output impedance, which is now inside the feedback
loop, won’t be ‘Zero Z’ but in our new project about 200 Ohm. Thus the tube
is inserted in both the signal path and Gainclone IC’s feedback loop.
Now we can introduce the Tubed Inverted Gainclone (not
really a ‘clone’ is it? I agree with Thorsten, but hey, this is how language
develops):
Here it is in all its glory:
(Updated November 2004 - now using T-Network solution. Honorable mention to
Franz Gysi.)
The tube I use in the JLTi amp and above, is the
Sovtek 6922, current production version. I’m
sure that others will have their own ideas. But a decent 6922 does work OK
at the voltages shown; in fact I’ve seen circuits using 6922s operating at
24V.
(But if you can get your mitty hands on a 6DJ8 Amperex Bugle Boy, don't hesitate!). That’s a bit low for my comfort but always check that there is no DC on
the input grid as grid current can be a problem, but not with up-to-spec
6DJ8s and less likely with 6922s. Also be aware of microphonics. The Sovtek that I use has been perfect in this regard, over a large sample,
this is a good thing when you are manufacturing. Only one twin triode 6922
is needed, one half for each channel and 6.3V 360mA min regulated supply.
Now we come to another key feature:
Notice the 1n3* cap? Perhaps you have. It serves a
couple of purposes, the first being that at very high frequencies it keeps
the feedback path short (a la 47 Labs contention that feedback should be
short but kinda difficult in a SERIES feedback situ). The second reason, in
conjunction with 4K7 it becomes a low pass filter, or LPF. You could say this is
bandwidth limiting but there is more to it. It does tailor the HF response
so that it is minus 1.25dB at 20KHz. The target response achieved by
listening test is shown here:
Caption: This is an actual measurement using an MLS
signal of my JLTi amp. Note dB scale on the left.
It has been noted that the original Gaincard, while
having great clarity and some have even said purity, it is known to be a
little relentless in the top end (a bit hot or glary), some of which is
claimed to be ameliorated by being left ‘on’ or burnt in. In the case of the
December 2001 Stereophile review, even a month wasn’t quite enough. It was
also suggested that it should be carefully system matched for optimum
results. Keep your sources warm, or speakers? Choose your cables with care
etc.
(Gainclones don’t like a lot of capacitance and the 0R22 Resistor
shown on the
Speaker Output helps to cope with moderate amounts - you may wish not to use it, but it's a safe option
keeping it in. With dynamic speakers, low capacitance cables,
like single core that are a bit inductive, try without the 0R22 Resistor).
I was also able to confirm these generalizations
myself, even with the Inverted Gainclone. One observer has described the
Gaincards and Gainclones as having a 'nervous' sound,. Very detailed but
also it bit on the edge. It became a kind of quest to sort
out what was happening and IF anything could be done about it. Happily I can
affirm something can be done and the end result is a less critical, more
balanced warmth sound.
Happily, it also, as it turns out, this is one aspect that can be tuned
by the individual DIY’ourselfer.
I suspected it was a slew-rate type problem, in which
case adjusting bandwidth passively by ear was the only way to go. What is the cause? It’s surely got to be feedback vs
phase problem, or in reality positive feedback because the feedback is
not keeping up pace. At some point the HF will start to lag beyond 90°, our feedback loop is not able to keep pace. An old but good
solution is following this guide: At the point where phase lags behind by
more than 90°, the open loop gain must not be higher than the closed
loop gain.
This is the NS data
sheet gain vs. phase response
Looking at it positively, this is actually not bad
news because we now can work on a mechanism by which we can
improve the sound of our Gainclones, tubed or not.
The graph indicates that an open-loop gain of about 30dB @ 200KHz. Above that the phase lag goes gradually off, beyond the 'ideal' 90°. It basically slows down. When feedback is not fast enough and in an Opamp IC, then slew rate
distortion results. That is why many solid-state amp design have a passive
low-pass filter (usually a full band-pass filter) right on the input stage.
So in that sense what I’m doing is not new, but what is new is this,
adjusting this low-pass by ear. It is very audible mechanism.
Bottom line, it is clearly born out by
listening tests, whatever the theory!!!!
In order for the final value (my choice) to be verified further by
ear, I asked a friend to take the amp for a period of time and adjust the
value. No other changes were to be made except this cap value. His system
has what I would call an even balance and no tendency towards either
brightness or dullness. I put a 390pF cap in there first and gave him a
handful of others with increasing values, one of which was the previously
preferred 1n3. This value was preferred as well by my independent tester...
My suggestion is that any DIY builders should also try
out different values and I would be interested in the feedback. Believe me,
it does affect the final sonics and it’s one of the things that are NOW
tunable for individual experimentation.
The NON-Tube version also did incorporate this kind of
approach, but it actually more complicated as it couldn't be stabilised. This and other matters like
keeping feedback more constant when we still have a pot directly in circuit. The article is no longer on the website, go for the buffer, it's worth the effort.
Join Us On
The DIY AUDIO FORUM
< <
BACK TO DIY MAIN PAGE |
