ONLINE - ISSUE 19
Lynn Olson's 2004
European Triode Festival Presentation, part 2
Re-printed courtesy of PFO Senior Assistant Editor Lynn Olson, and are re-printed from the Nutshell Hi-Fi site at http://www.nutshellhifi.com/library/ETF.html.
Here's the graphic you saw on the first page, minus the dramatic gesture. But there is an important point here, so a little drama is called for. For years, amplifier designers have assumed speakers are perfect, and speaker designers assumed amps were perfect. But we don't live in a Platonic world of perfection, but the real world where things are far from ideal.
In reality, speakers (and I mean all speakers here, none excepted) store quite significant amounts of resonant energy, and for relatively long periods of time compared to the transit times within an amplifier (tens to hundreds of milliseconds). The Q's are high, and there are scores, or even hundreds of resonances. Perhaps worst of all, the speakers we like have efficient coupling between the amplifier and the moving elements (high BL factor), so these resonances are transmitted more efficiently back to the amplifier.
If our output stage was distortionless (which speaker designers assume), well, we wouldn't care very much. But unfortunately our output stage is not only not distortionless, but it typically has the most distortion of any stage of the amplifier. So the worst stage of the amplifier is in intimate contact with the most massive and worst-behaved transducer in the entire audio chain.
You can now see the potential for trouble. We were previously concerned about the audibility of the B+ filter cap in the current-loop signal path; this is much worse, unwanted speaker resonances crossmodulating with the desired signal in the forward path. Again, all the nonlinear elements contribute to the distortion, this time a crossmodulation term between the back-EMF of the speaker driver and the desired audio signal.
Amplifier designers who think they are smart will come up with simple LCR networks that simulate the overall impedance of the speaker; unfortunately, simple networks like this assume the drivers themselves are perfectly flat (and distortionless) within their working band. Real drivers have not just one, but many many resonances, particularly at the upper part of the working range. These resonances appear at the plates of the output stage, multiplied 20 to 30 times by the output transformer.
This, by the way, is why I favor output stages that have a very smooth output impedance as as the waveform swings through the duty cycle - an area where Class A PP DHT triodes excel, and for the first several hundred milliwatts, SE DHT triodes are pretty good too. We get in trouble when there are "lumps" in the dynamic output impedance, as in Class AB or pentode operation. This can greatly magnify audibility of speaker coloration, with the worst case being Class AB transistor operation - this is why transistors and horns are not compatible. Horns have complex and extensive back-EMF resonances, combined with very low IM distortion (so amplifier distortion is much more audible). Although the feedback loop in the transistor amp tries its best to simulate a low-value resistor in the output stage, the simulation is not that effective at removing the high-order terms from the transistor conduction characteristic—and with low IM distortion speakers, very audible.
By way of example, the upper pair of charts are a commercial speaker I measured a few years ago on MLSSA, and the lower pair of charts are the Ariels. I don't mean to hurt anyone's feelings, but horn speakers usually measure worse than the first pair of charts. Although the time response of high-efficiency speakers is problematic (due to reflections from the edge of the horn mouth), there's an important compensation: the IM distortion is typically ten times better than direct-radiators. So horns will favor some types of music and not favor others.
In general, efficiency and low IM distortion come at the price of more resonant impulse response. Not the leading edge of the impulse, which only affects the high-frequency limit, but the stored energy on the trailing edge, which affects impressions of speaker coloration and resulting back-EMF resonances fed back to the amplifier.
To decrease the delayed energy of horns (again, this is not related to linear-phase crossovers or the leading edge of the impulse), treatments to "smooth out" the hard edge of the horn (Tractrix profiles, UREI-style foam or felt strips) can be remarkably effective at only 1dB cost in efficiency. Going further, some prototype horns have acoustically transparent diaphragms, such as horn-electrostats or even more exotic, horn-ionic speakers. Of course, that's a topic for a whole different talk ...
99.9% of all transistor amps have the output topology shown below - the most expensive high-priced amps have lots more transistors, each wandering around thermally and finding it's own Class AB transition point, and all of the multiple loops going through banks of "computer-grade" electrolytics, sometimes as large as 120,000uF. The circuit is only "direct-coupled" if you ignore the high-noise, high-coloration power supply—which is usually conveniently drawn on a separate part of the schematic. In the real world, the plus and minus power-supply leads radiate rectifier hash and Class AB switching pulses into the circuit board.
Speaker designers abandoned electrolytic caps back in 1980. But as you can see, they are in almost every transistor output stage. The only way around this is very rare - full regulation for the entire amplifier, not just the inputs and drivers. It's the outputs that need the regulation and isolation from the banks of electrolytics, but this is almost never done. The only transistor amplifier that I know that is fully regulated is the R.E. Designs LNPA-150—which, by the way, doesn't sound like a transistor amp, but more like a very good triode amp.
In conclusion, the current loops in this picture why transistor amps sound the way they do, and why they don't "play nice" with reactive speakers. Groups of not-very-well matched transistors are switching on and off at 0.7V signal levels, horrible-sounding electrolytics are in the direct signal path (the current loop, remember), unfiltered (filters require a RC or LC term, not just C) power-supply hash appears directly on the rails, and the feedback network is supposed to magically correct for all of this. Of course, it doesn't. The distortion meter is fooled - with a nice resistive load, of course—but the ear is not. And the more transparent and revealing the speaker, the worse an amplifier like this will sound.
For an amplifier to sound good with a transparent speaker, the intrinsic distortion of all the parts in the current loop has to be as low as possible—not just by measurement, but subjectively as well. We don't have a good handle on why caps don't sound very good, but the sonics are clearly audible, and not to be ignored simply because DA and DF doesn't appear to tell the whole story.
Text © Lynn Olson 2004. All Rights Reserved.