Do you still need matched tubes if running a CCS ?

[Nihilist]

However you do it you still have the problem that a CCS controls the average current, but you should be controlling the quiescent current. These two are only the same for a perfectly linear valve. Second-order distortion, present in all valves, increases the ratio of average/quiescent current. No problem for a small signal stage, but important in an output stage. Even traditional resistor cathode bias has to compensate to some extent for this, by running with a higher than optimum zero-signal current so the shift in bias with a signal pulls the quiescent current back nearer the correct value. Fixed bias avoids this problem. CCS bias makes it worse.

There is another problem with CCS bias. A bypassed CCS behaves at low frequencies quite differently from a bypassed resistor. This can cause LF instability if global feedback is used, and LF phase shift with no feedback. I would never use CCS bias for an output stage. If you want to match currents, then use a servo which samples at zero-crossing only.

I'm not sure, as I'm still learning abbout this stuff, but isn't all the stuff you just said irrelevant if you are running the output stage only Class A ?

As far as I know, that is the only way this circuit is to be run.

As I said, I could very well be wrong.


.......................Blake

 

[DF96]

It is still relevant to Class A outputs, although Class B is worse in this respect. The bias shift away from what you set with zero signal means you either accept more distortion and less ouput, or you compensate by sitting off the correct bias for small signals but slide back for big signals. I first saw this issue discussed by Mullard in connection with biassing EL84 for the 5-10 amplifier. They gave two different cathode resistor values. The higher value (i.e. lower no-signal current) gave lower distortion for music, but could not handle continuous sine wave testing as the bias shift was too great.

 

[Michael Koster]

There's another refinement of the CCS+Bypass circuit, which is to use 2 electros in series, opposite polarity, between the cathodes. Then apply a high impedance bias voltage to the point the 2 caps connect together. (maybe a 100K resistor to ground). This DC voltage across the caps will prevent low level signal rectification by the electrolytic caps, if that's a concern.

The other problem is related to how much voltage shift at the cathodes full signal vs. idle. For this to work well, the load lines should be chosen for good symmetry (current swing +/-). High flat load line, less efficient but lower 2f distortion.

Class AB is not an option without additional tricks e.g. cathove voltage clamp or DC restorer type circuit.

Michael

 

[Shoog]

I bias my cathode nodes with a 1meg resistor.
I would like to here some feedback from those who have actually tried this as all I have heard from those who have is very positive. I am suspect of theoretical problems which haven't been tested in the field. Bass response has certainly not presented itself as an issue.

Shoog

 

[DF96]

I remember advising someone, either here or another forum, only a few months ago. He had taken a working design and 'improved' it by putting CCS bias into the output. The result was LF instability. Definitely not a "theoretical problem" which hasn't happened in the field. Real amplifiers always obey Bode-Nyquist stability theory, whether their designer believes it or not.

The LF problem is that a bypassed resistor is a lead-lag network, so no phase shift at low frequencies. A bypassed CCS is a high pass filter so 90 degrees phase shift at low frequencies. This can make a huge difference.

 

[smoking-amp]

This issue of the tube nonlinearity affecting the optimum tail current or tube biasing came up before for CCS tail/anti-triode schemes. One could use a high value resistor from the cathode(s) to the CCS(s) gate(s) to control the "CCS" current(s) slightly. This then becomes equivalent to a finite impedance at the tail. Probably similar to the tail Z required to null 3rd harmonic dist. Then there was the idea of putting a LP filter in this control path so the "CCS" would only follow the envelope. (which would actually be necessary for the anti-triode scheme, to avoid removing the triode side "signature")

 

[Shoog]

I never use global feedback, only local, so the issue has never arisen in any of my designs. Would I be correct in stating that the phase shift is a product of undersized bypass caps - I have always advocated using at least 1000uf:1000uf (series value 500uf) to avoid this.

As I have stated before I use CCS for the very specific purpose of perfect current matching in Torodial outputs. They have always worked fine for me in this application. If using EI's then the requirement it relaxed somewhat, but current matching is still of benefit. Unless you are the sort of person who enjoys bias adjustment on a regular basis, any other biasing arrangement in a PP amp will have rising distortion with time - for me the lack of a need for adjustment (ever) holds a lot of value.
I have also used Garter bias in the same position, with cathode to cathode bypassing - and this has worked very well where the wasted voltage is not an issue.

Shoog

 

[Shoog]

I think it was Tubelab's who made the observation that in a desperate attempt to avoid using electro's as cathode bypass caps, most people were using very marginally spec'd film caps. This had the effect of increasing impedance at low frequency, and introducing phase shift.
It seemed that the cure was worse than the disease. It was advocated to use bypass caps in the thousands to avoid this issue.

If anyone else can remember the exact thread and confirm my memory, I would be interested.

Shoog

 

[piano3]

I have used the schematic as posted-but with very different CCS's- by Michael Koster with a PIO cap of only 20uF and simply regard the cap as two 40uF in series with the centre as AC ground which is then disconnected. I actually think the value is a bit high. As far as I understand it,not only the interaction of rk with the cap is relevant but the tuned circuit formed by the cap and the primary inductance of the OPT with the Q damped by the tube's dynamic resistance.

I will be trying Shoog's idea on the next build as I am totally taken with the topology!

 

[piano3]

....and if it is thought that it is difficult to arrive at a reasonably accurate bass response then;

find a slightly warped LP, increase cathode coupling cap until the warp creates an unacceptable disturbance and then back off.

In my previous post,when I say that 20uF is a bit high it is in the context of L primary being more than 100H and rk being little over 150 ohms

 

[DF96]

The phase shift will always happen. The size of the bypass cap determines where it happens. Roughly speaking LF phase shift will start when the cap impedance is equal to 1/gm. For a CCS the phase goes up to 90 degrees and stays there right down to the point where the cap impedance is equal to the CCS impedance - seriously subsonic! For a resistor (or a deliberately degraded CCS) the phase starts returning to zero where the cap impedance is equal to the resistor - typically only 3-5 times lower freq so the phase never reaches 90.

So, ballpark figures: gm=10mA/V, so 1/gm=100R, C=500uF, CCS=1M, Rk=500R. Phase shift will be about 45 degrees at 6Hz. For CCS phase will stay high until 0.0006Hz. For resistor phase will be back to 45 degrees at about 1.2Hz. If you use much smaller capacitors, as many do, then you can easily see that significant phase shift starts well within the audio band. With global NFB you could easily get an LF peak right in the record warp region.

So OK, I will modify my advice: don't use a CCS in the output unless you intend to use some serious capacitors too!

 

[bigwill]

I remember advising someone, either here or another forum, only a few months ago. He had taken a working design and 'improved' it by putting CCS bias into the output. The result was LF instability. Definitely not a "theoretical problem" which hasn't happened in the field. Real amplifiers always obey Bode-Nyquist stability theory, whether their designer believes it or not.

The LF problem is that a bypassed resistor is a lead-lag network, so no phase shift at low frequencies. A bypassed CCS is a high pass filter so 90 degrees phase shift at low frequencies. This can make a huge difference.

'Twas me.

I won't be doing it again! Next time, I will be using fixed bias with a clipping bias servo, carefully designed so its time constant is way below the LF rolloff of the amp.

 

[Shoog]

So OK, I will modify my advice: don't use a CCS in the output unless you intend to use some serious capacitors too!

Thanks for the correction - that is what I have always advocated from the start.

The question is does the penalty of using big electro caps (suitably bypassed in the critical high frequency range) outway the gains of differential performance. My experience is that it certainly does not.
A servo maybe the best option, but they are quite a bit more complicated to implement.

Shoog

 

[richwalters]

So, ballpark figures: gm=10mA/V, so 1/gm=100R, C=500uF, CCS=1M, Rk=500R. Phase shift will be about 45 degrees at 6Hz. For CCS phase will stay high until 0.0006Hz. For resistor phase will be back to 45 degrees at about 1.2Hz. If you use much smaller capacitors, as many do, then you can easily see that significant phase shift starts well within the audio band. With global NFB you could easily get an LF peak right in the record warp region.

So OK, I will modify my advice: don't use a CCS in the output unless you intend to use some serious capacitors too!

Well written. Excellent.. Youv'e got it. One can't always "CCS it", with flighty tubes, high gm and wide b/w, an ultrasonic oscillator is easily created.

Going a tad further, those examining push-pull Williamson concepts notice the increase in LF response well below o/p tranny cutoff when the loop gain versus frequency runs out; this is the Achilles heel stability issue, je higher global nfb quantity the worse this sub hump will be. Radiotron handbook 4th ed p.256 and p.751.
The characteristic hump response shown in attachment. It can be reduced by shunting each of the input coupling caps of the o/p stage by an electrolytic uF plus a res (typ 10uF+1M) which will alter the low end response. These have to be calculated. It depends on the b/w of the op tranny, loop gain and the other interstage coupling caps.
High power amps with the combination of wide bandwidth and high global nfb (20-30dB) are very prone to subsonic "lockup" which hammers the o/p tubes with high currents, i.e, o/p tranny Bsat is easily reached with subsonic frequencies with low AC excitation op/tranny primary volts. In effect, the classic formula, Ae = (L x I/Bmax x N). The Core area simply becomes too small for the subsonic signal volts as iron permeability runs out, slamming primary Z.
I use an 18dB HP subsonic filter of 15-20Hz on the input stage. Doing this, the power handling ability with "rough" input signals i.e outdoor PA with wind is near perfect.
Examining the open loop characteristic of amplifiers, another solution to reduce the LF hump is to reduce the global nfb, which will proportionally raise the s/n ratio and thd and worsen effective loudspeaker damping. It all depends how much medicine one applies.

Morgan Jones Book, Valve amplifiers 4th Ed doesn't cover this topic in detail, p.414 is a descriptive hint and I hope the next edition covers a bit more of this detail.

This whole subject of Bode plots, critical circuit damping with optimum LF impulse transient response is a complex subject but extremely important one sonically.


richy


(M.J remember all the bumpf I sent you ??)

 

[piano3]

Hello DF96,

I do not use feedback at all; in each cathode CCS impedance at least 50M and possibly more than 100M , cathodes coupled by 20uF. No audible problem except that record warps are more obvious than in previous SE amplifier. Would you still say that it is likely that a problem is there even though neither myself or any musician colleagues who have heard the amp so far can hear any or do you think that the very high sink impedance compensates for the low value of C?

Actually, I need to think long and hard about what you have already said!

 

[DF96]

With no feedback you won't get an LF peak. A common CCS (i.e. shared by the push-pull pair) needs no bypass. Separate CCS, one per output valve, must be bypassed. The LF corner frequency is then f = gm / ( 2 pi C ), so 20uF seems to be far too small. For gm=10mA/V f=80Hz. OK, I suppose small speakers won't have much output below 60Hz anyway so unless you are listening to organ pedals or reggae you might not miss much. I suppose another problem is that big capacitors tend to be low tolerance (10%, 20%?) so the two channels could have phase differences which will affect stereo imaging. You can plug in your output gm to get a more accurate figure.

The high CCS impedance does not compensate for low C.

 

[piano3]

The speakers are Tannoy Chatsworth and indeed low organ notes are reproduced very well-about this there is no doubt whatsoever! Is it not possible that what I suggested about an interaction between various time constants, including a somewhat damped resonance between the OPT primary L and the cap is indeed the case? To clarify, the cathode coupling cap is a single cap and not two with the junction connected to earth via a resistor. I must admit,DF96, that after reading your very clearly expressed posts I am somewhat at a loss to explain what I am hearing!

 

[piano3]

One detail I omitted(if it makes any difference!); the CCS's are from a -ve supply.

 

[piano3]

o.k. some simple figures; 100H, 10uF resonant peak 5Hz,less than 2 octaves below lowest tones of acoustic instruments, requiring 6.3 kohms for critical damping, 100H, 1000uF resonant peak 0.5Hz,more than 3 octaves lower, only needing 630 ohms for critical damping. Surely this is relevant?

 

[DF96]

If a single cap joins the two P-P cathodes then the LF corner frequency is halved, as you can treat it as two caps each of double the value. This assumes that the phase splitter operates down to DC, or at least well below the cathode corner frequency. That might explain things. Sorry, you did say this in an earlier post but I didn't spot it.

The OPT primary sees the cathode bypass capacitor indirectly, as there is a valve in the way. Roughly speaking, the anode impedance has added in series with it the transformed capacitor reactance - reactance multiplied by mu. Or capacitor value divided by mu. The net result is likely to be less LF, but it depends on the details. There could be an LF peak followed by a quicker roll-off, especially with a triode output.