slew rate redoux
[Snipped from audioasylum.com]
Posted by Lynn Olson (M) on June 03, 2002 at 11:37:45
In Reply to: Confessions of a plate choke convert & a few CCS questions (long) posted by Doug Flynn on June 03, 2002 at 02:41:03:
One of the traps people fall into is not realizing just how low DHT distortion really is … and that DHT’s need a not only a lot of voltage, but enough current to overcome a substantial amount of capacitance (this controls slew rate).
How low? Well, the Amity has no trouble putting out 16 PP watts at 0.3% distortion, and 1.6 watts at 0.07%. That’s *including* the distortion of a PP 7119 driver. If you really went nuts on the driver and used something like another 300B (!) as a driver, I’d bet the distortion might fall even more, maybe down to 0.2% or less.
300B’s need anything from 42 to 60V rms of extremely linear drive to realize their full potential. I doubt that most people have even heard what a 300B really sounds like … many DHT amps have more driver distortion than the final stage, and the drivers have the additional disadvantage of not delivering enough current to the DHT grid.
Although bare-bones engineering indicates that only a few milliamps are needed for full power at 20kHz, it’s easy to forget that slew distortion is merely the current equivalent of voltage clipping … and we all know distortion doesn’t just go from zero to full. Instead, distortion gradually falls with decreasing level. The same thing happens with *current* distortion – the less current demanded, the more linear the driver is. Well, we can’t do much about the current demand of the 300B grid, but we can certainly increase the current running through the driver … by several times if transformer, choke, or active loading is employed.
Improving the slew rate by several times doesn’t sound like what you’d expect. People expect shimmery triangles and amazing transients, but that’s not what you notice. Instead, there is a dramatic reduction in the “electronic” coloration of the sound – in fact, most of the “tubey” vintage aspect of the sound disappears.
Since slew distortion is greatest at the zero crossing of the waveform, anything we can do to remove even the slightest trace of slew distortion (which is a different animal than generic THD) will give an astonishing improvement in hearing subtle details in complex passages. After all, we can all live with some distortion on the signal peaks, but we *don’t* want any distortion at all in the zero-crossing region.
Personally, I think it’s impossible to overdesign a driver or power supply. You want the most linearity possible with lots of linear current available, and the power supply needs to deliver lots of peak current while isolating the audio circuits from rectifier switch-noise and AC line noise.
If you’re tired of vintage sound, improve the driver. Go nuts. Aim for 3 to 6dB of headroom or more, at least 15-20 mA of operating current, and transformer, choke, or active loading. (And yes, they all sound different!)
——————————————————————————–
Re: Confessions of a plate choke convert & a few CCS questions (long) – Ken Gilbert 07:11:52 06/04/02 (5)
In Reply to: Re: Confessions of a plate choke convert & a few CCS questions (long) posted by Lynn Olson on June 03, 2002 at 11:37:45:
“… and we all know distortion doesn’t just go from zero to full.”
lynn,
your observations wrt slew rate limiting are not exactly what i have found, namely in that slew rate limiting is a gradual effect; instead i have found that either a stage is SRL or it is not… either there is enough current to drive the downstream capacitance, or there isn’t. when there is enough current, you aren’t SRL. this is distinctly different from the increase in THD that comes from increased signal level output, which is as you’ve described gradual in onset, but always there, even at the tiniest of levels. in contrast, below the max slew rate, there is ZERO SRL.
now, i agree that music is quite dynamic in nature and the requisite slew rate will therefore change throughout the performance, so the effect of SRL may come and go as the program material changes, but that’s why you account for a cushion of headroom in your calculations for required SR.
ken
——————————————————————————–
Partial slew rate distortion – Lynn Olson 01:36:01 06/06/02 (4)
In Reply to: Re: Confessions of a plate choke convert & a few CCS questions (long) posted by Ken Gilbert on June 04, 2002 at 07:11:52:
Hi Ken, just read your discourse on your Web site, but I must respectfully disagree about the all-or-nothing nature of slew rate distortion. I’ve been down this path back at Audionics in the late Seventies with solid-state amplification.
What we saw on the scope and spectrum analyzer accorded with the final voltage-gain stage gradually losing its linearity as more and more current was demanded by the dominant pole of the amplifier, then reaching a “hard” slew limit when all the current was finally used-up by the capacitive load. And yes, feedback certainly worsened the slew limiting, with the worst-case situation being reactive (speaker) loads that ate into the available phase margin. The aspect of the CC-2 amplifier that the late Bob Sickler was most proud of was the combination of generous phase margin, high slew rate, and low intrinsic distortion in the forward path. This resulted in fast settling times even with highly reactive loads, and unexpectedly, much improved Class AB crossover transitions due to adequate phase margin with any load.
Moving on to vacuum-tubes, it’s obvious from inspection of the published plate curves that the slope of the load-line affects distortion, and the opening-up of a flat (resistive) load-line into an ellipse (partly reactive) will make distortion significantly worse. When the elliptical load-line swings through the low-current region, abundant upper harmonics are generated.
This is a progressive effect, with distortion increasing with signal level, slope of the load-line, *and* percentage of reactance in the load. What’s unfortunate about capacitive (as opposed to inductive) loading is that it falls in a region where the ear is particular sensitive to distortion – distortion audibility roughly parallels the Fletcher-Munson curve, so IM distortion at 5kHz is 20-30dB more audible than distortion at 50 Hz.
A few pF here and there is hardly a concern for transistor circuits that cheerfully operate at 100mA or more, but is a big deal for vacuum-tube circuits that operate at a tenth of that. The “stray pF’s” are very hard to avoid: there’s the Miller capacitance of the power-tube grid, the stray C of the choke, transformer, or dynamic load, and the inevitable capacitance of the socket and wiring itself. So our driver circuit is always going to see a net load of 50 to 100pF or more, and has the additional awkward requirement of swinging 40 to 60V rms per grid – at distortion figures below the already low DHT figures! This is not a trivial requirement for any driver section – trivial perhaps at 1 kHz, but not so easy at 20 or 50 kHz.
Interestingly enough, conventional RC-coupled PP amplifiers drive each power tube with what amounts to individual SE drive – one driver per output tube, of course. This means the elliptical load-line of the (reactive) power tube is presented to a SE driver circuit.
This is the subtle reason I chose to experiment with the little-explored and archaic path of IT-coupled PP drivers and PP outputs. I wanted to sum the PP drivers in the interstage transformer, so the grid-lines would be straight and parallel (PP Class A operation) instead of the highly curved grid-lines of SE operation. When grid-lines are straight and parallel, a change in load from resistive to reactive has very little effect on distortion. In SE operation, distortion typically increases several-fold when a reactive load is encountered.
The net effect of this choice is the PP 300B grids are presented with a drive that’s low-distortion even at the highest frequencies, something that would not be true if conventional RC-coupling was employed. With pentode output tubes, a reduction in driver distortion is much less important, partly due to more moderate driver requirements (in both voltage and capacitance), but more importantly, the substantially higher distortion of pentodes (even triode-connected pentodes) compared to DHT’s.
I admit this concentration on maximizing linearity at high frequencies is well off the mainstream – it’s partly a result of my time at Audionics and Tektronix, but also partly the bias of a speaker designer that’s forced to pick among very imperfect drivers with no recourse to feedback. You can use feedback with moderate success at low frequencies in loudspeakers, but things really fall apart at high frequencies, making feedback impossible to stabilize above 200 to 300 Hz thanks to cone break-up and the difficulty of finding an appropriate sensing point.
In speaker design, the entire speaker is made or broken in the critical 1 to 5 kHz region, the region of peak sensitivity for the ear. I carry this preference forward into electronics, and focus on HF performance into complex loads (which both speakers and triodes certainly are).
This is a broad definition of slewing – current distortion into a capacitive load – but I feel it’s a useful distinction from midband distortion, where by definition the circuit is operating well away from edge-of-band inductive and capacitive loading. Just as voltage distortion is not an all-or-nothing phenomenon, neither is current (or load-dependent) distortion. As more and more of the quiescent current in the tube is consumed (or reflected) by the load, the amount of current flowing through the tube becomes a larger and larger percentage of the quiescent value. The plate curves of most triodes are well-behaved at high currents, but get a lot less linear when the current flow drops to 10-20% of the nominal quiescent value.
Another undesirable side-effect of the capacitive load is the phase-shifting of energy … the capacitance reflects essentially all of the current back to the plate. This 90-degree phase-shift results in the region of maximum current demand falling not the top of the waveform, where you’d expect it, but the zero-crossing region instead. Since audio is a logarithmic perception, tiny degradations 20 to 60dB down from the loudest signal are important and perceptible – and the most likely place to find these small-but-significant signals is in the zero-crossing region. This is why a tiny amount of corrosion in a dry circuit can have devastating impacts on perception of space, dimension, and emotional subtleties – the damage is all occuring in the worst possible portion of the signal swing.
Hard-to-measure bursts of distortion that occur around the zero-crossing region have a subjective impact entirely out of proportion to the averaged-THD numbers. Some of the causes are slew distortion, Class AB crossover distortion, PCM errors around the LSB, or minor corrosion effects in a dry circuit. The reason is simple: although the burst of distortion may be brief as an overall percentage, it is selectively masking or actually removing low-level information from the signal.
Aloha Audio
——————————————————————————–
Re: Partial slew rate distortion – Ken Gilbert 11:14:49 06/06/02 (3)
In Reply to: Partial slew rate distortion posted by Lynn Olson on June 06, 2002 at 01:36:01:
thank you for the very detailed and thought-provoking post, lynn.
i appreciate your comments and am re-thinking my own views on SRL and distortion; i think your point about greater and greater proportions of the signal current being “used up” so to speak by the reactive load impedance is very salient.
even though below the SRL we don’t have a situation where the tube is running out of current (as in zero left to give), we DO have the situation where even if the plate is loaded by a perfect CCS the downstream load is causing drastic current swings, which will of course affect the stage’s linearity–at least this is how i interpreted the main jist of your post. as the current swings drawn by the load are increased, the beautifully flat horizontal AC loadline of the tube and CCS alone gets tilted more and more vertically, venturing down into the curved portions of the curves and adding THD.
along those lines 🙂 your other points about the PP driver to PP output through IT are very keen as well–PP class A is the best point/topology we’ve got to contend with reactive loads (wide current/voltage swings), and indeed the output grids (especially of triodes) can be quite reactive, especially as we push the HF -3dB limits out to 50, 100, or even 200kHz in search of maximal flatness in the audio passband.
again, thanks for taking the time to respond. i will most likely be updating my website with a few of these caveats! 🙂
ken
——————————————————————————–
Re: Partial slew rate distortion – Lynn Olson 12:44:55 06/06/02 (2)
In Reply to: Re: Partial slew rate distortion posted by ken gilbert on June 06, 2002 at 11:14:49:
… “as the current swings drawn by the load are increased, the beautifully flat horizontal AC loadline of the tube and CCS alone gets tilted more and more vertically, venturing down into the curved portions of the curves and adding THD” …
Actually, it’s worse than the load-line getting slanted down from the horizontal (no current fluctuation, tube is looking into an infinite impedance). A slanted load-line simply represents a finite resistive load for the tube, usually expressed as a ratio to Rp (dynamic plate impedance).
(Minor Digression: Traditionally, 3.5~5 times Rp was considered a pretty easy load for the tube, keeping in mind that removing the cathode-bypass cap also multiplies the Rp, thus both generating local feedback (good) but also degrading the load/Rp ratio (not good at all). When the load/Rp ratio is unfavorable (1:1 or worse) local feedback can’t correct the descent of the load-line into the ugly low-current region, where abundant upper harmonics are generated. Local or global feedback will reduce distortion in direct proportion to the reduction in gain, but has no effect on the proportion of harmonics. So whether or not local feedback, or degeneration as it is sometimes called, is used, it can’t change the behavior of the load-line, which in turn controls the proportion of upper harmonics. The relation of the load-line to the grid-lines creates the tube’s distortion “signature”.)
The proportion of upper to lower harmonics is critical to the sound of electronics. In the late Fifties, there were serious proposals by Norman Crowhurst, D.E.L. Shorter of the BBC, and other engineers to weight the harmonics by the square, or preferably, the cube of the order in order to better correlate with the subjective impression. This was never taken up, partly due to marketing reasons, and partly due to the extra complexity of measuring each harmonic at least out to the 10th and calculating it’s impact (this was back in the day of the slide rule, remember). But what they said then still applies now, especially when you consider the devastating impact of high-order terms on IM distortion. D.E.L. Shorter mathematically demonstrated in a series of “Wireless World” articles that when you have 3 dominant tones of roughly equal magnitude, IM sum-and-difference terms start to outnumber THD harmonics. When you have 4 dominant tones, then IM distortion greatly dominates.
In other words, the more spectrally complex the source material is, the more IM sum-and-difference tones you get … and this goes up as the *square* of the number of tones present in the original source. So with complex material, you’re not really hearing THD, but thousands of sum-and-difference IM distortion components. This is much worse when you start with high-order THD … the primary benefit of low-order (2nd and 3rd) harmonics is they generated fewer IM sidebands, thus giving a cleaner spectrum overall. The reason a massed chorus becomes a roaring haze of distortion, for example, is the very dense spectrum generates thousands of sidetones that were never present in the live, unamplified performance, and ear instantly detects the fraud. That’s why you can tell “PA” sound at such a great distance.
Returning our old friend the load-line, yes, a resistive load creates a slope, and we have to watch out as the load-line approach the low-current region. But a *reactance* – either inductive or capacitive – creates an ellipse, and ultimately, a full circle when the load is 100% reactive with no resistive component. (A circular load-line is what blows out bipolar transistors when the circle momentarily exceeds the Safe Operating Area of the transistor current/voltage curve. It only takes a few milliseconds of exceeding SOA to melt down the tiny gold wires going to the chip and destroy the device. This is why you see fast-acting current-sensing protection circuits in bipolar-transistor amplifiers.)
Fortunately in the tube world, we don’t need to worry about a circular or elliptical load-line destroying the device. It takes a prolonged thermal overload to wreck the plate structure, and even voltage flash-over can be tolerated occasionally. So outright destruction is not an immediate concern.
But an elliptical load-line *does* edge us closer to that troublesome low-current region, and also spells big trouble for pentodes that like to see a defined resistive load for lowest distortion. (Pentodes give lowest distortion not into an infinite load, but a specified resistive load. If you really want to see pentode distortion skyrocket, just add a little capacitance to that ideal resistive load. Look at the wavy plate curves, and the reason becomes obvious.)
In the output stage, of course, we are in a worst-case situation, driving a messy mechanical motor-drive chock-full of resonances. We get not just the expected resonances at driver resonance (Fs), but lots of other resonances thanks to cone standing waves, internal cabinet standing waves, spider resonances (no relation to the movie), dust-cap resonances, standing-waves between the horn-mouth and the reflective phase plug/waveguide, etc. etc. etc. Not only that, but our favorite speakers, the ones with the lowest IM distortion (horns and electrostats), are the worst behaved from the viewpoint of internal resonances that are reflected back to the amplifier (also known as Back-EMF). Direct-radiators are pretty ill-behaved too, thanks to lots of stored energy in cabinets and standing-waves on the cone itself.
All this trash gets sent right back to the amplifier output stage, which has to play music right through these harmonically unrelated back-EMF resonances. In the real-world amplifiers we can buy, though, the amplifier does interact with the back-EMF resonances, exaggerating the audibility of what might otherwise be a fairly minor speaker coloration.
So we care about behavior with reactive loads, no matter whether it’s in an output stage or in an intermediate stage. It’s probably not as important in an input or preamp stage, simply because the voltage/current swings are so much less and the loads are well-known and well-defined.
Aloha Audio
——————————————————————————–
Re: Partial slew rate distortion – Lynn Olson 13:16:23 06/06/02 (1)
In Reply to: Re: Partial slew rate distortion posted by Lynn Olson on June 06, 2002 at 12:44:55:
Quick Note: So far we’ve been discussing straight (resistive) and elliptical (reactive) load-lines. This assumes distortionless loads! In reality, the lines themselves are kinked or bent a little … with a transformer, for example, things get really weird as we approach 20 Hz. Our pretty little ellipse actually gets dented, something no tube can correct for. The presence of DC, and the choice of core material, have a big, and clearly audible, effect on the linearity of transformers and chokes.
Don’t think we’re home free with active loads either; slew rate considerations *do* apply for current sources, so even fancy active loads can add unexpected colorations. Current sources sound quite different from each other, and who knows which one is “right.” Those tiny little semiconductor pF’s are a lot more audible in the midrange than you’d think. (As capacitors go, bipolar and MOSFET transistors are very low quality, with base/gate capacitance modulated by current, voltage, and temperature.)
So even a dumb load-line has slope, elliptical reactance, and nonlinear distortion (which in turn is frequency-dependent). All this stuff is cross-multiplied by the plate curves of the tube itself. Just think, all this in one tube stage!
——————————————————————————–
thanks again lynn (NT) – Ken Gilbert 11:59:37 06/07/02 (0)
In Reply to: Re: Partial slew rate distortion posted by Lynn Olson on June 06, 2002 at 13:16:23:
