Analysis Paralysis to the max is where we are at!
Quickly jotting down a list of thoughts and findings since my last post on this topic:
Full-range drivers likely to gain the most
Multi-way speakers are pre-emptively designed with various enhancements / compromises so as to avoid the problems associated with playing a wide range of frequencies through the same driver. So much so, that as soon as the base problem of intermodulation distortion (etc.) rears its head, the go-to solution is basically: don’t think, just increase the number of drivers from 2 to 3, or 3 to 4, ad infinitum.
There are clear benefits to reducing the bandwidth passing through each driver, and reducing the signal bandwidth passing through the amplifier. It’s great for sales, too! However, I’m sure that that last point has zero influence and people are purely guided by what’s sensible.
Conversely, there’s a different list of benefits of keeping the number of speaker drivers down to a minimum. Not to evangelize, but to my ears there’s a sort-of ‘chorus’ effect when multiple speakers attempt to play the same thing. Nevertheless, multiway speakers could all individually benefit from what’s being discussed, though probably to a smaller degree than with FR drivers.
What’s the best output impedance?
Recapping, there have been a number of papers, by Mills and Hawksford among others, describing how the distortion of dynamic coil and magnet type speakers can significantly improve when operated via current drive. However, things are not so simple.
Short answer: “it depends”.
High output impedance:
- Clear benefits for the upper frequencies of most dynamic transducers (magnet + coil), where stray inductance would otherwise produce IMD and hysteresis distortion.
- Neutral / no clear benefit in the middle region, where impedance is at a minimum.
- Likely detrimental in the bass, especially if the impedance has one or more prominent peaks from mechanical resonances.
Low output impedance:
- Worst case for high frequencies (e.g.: >1kHz for a typical 4-6″ full-range or mid-range driver)
- Neutral for the middle range where impedance is low.
- Likely best case for bass output.
Bass performance has been discussed extensively on the DIYAudio forum, where I sometimes chime in. A couple of insights, from memory:
- Voltage control stabilises the speaker output across the resonant peaks in the bass. It eliminates any need for notch filters, which could otherwise get messy due to drifting speaker parameters.
- For a typical overhung voice coil design, the magnetic gap “hogs voltage” in the bass resonance region, because the overhanging parts of the voice coil have a disproportionately lower reactance (as the magnetic field density diminishes). From this it seems that the total voltage across opposite ends of the the voice coil should have a fairly constant and linear relationship to the voltage across the magnetic gap.
- There was some argument that current control could compensate for thermal effects. However, thermal compression seems relatively benign (I’ve run speakers through incandescent lamps with surprisingly modest degradation of sound) and a big caveat is that thermal runaway could ensue. Voltage control actually helps protect speakers against catastrophic failure.
More recently, someone argued that a distortion component related to BI modulation at bass frequencies could be reduced with current drive because of the simpler equation using BI instead of BI².
However, I’m wary of extending current drive to the bass because of the added complexity of notch filters. Even in its simplest form, a single active notch filter can introduce noise and distortion. For a feedback-based topology, gain peaks have lower feedback and therefore tend to have higher distortion. Furthermore, a pre-amp stage that filters the power amplifier’s input can reduce the net gain, but not the distortion. The end decision on topology will have to consider the overall linearity, and linearity of the output stage in particular.
Topologies, topologies
I’m going with a “mixed mode” design. One that will have a low output impedance in the bass, and high output impedance at high frequencies.
Greed!
The trouble with a classic mixed-mode design, is that the transition between voltage feedback and current feedback is slow and gradual with a mere 6dB/octave slope. This obviously limits the potential performance at both frequency extremes. If we want a very high output impedance at high frequencies, the bass is likely to have a low-ish damping factor. Conversely, if we want a super-high damping factor, the treble will suffer.
What if we could have a fundamentally different design that has a steeper transition between the two? The dual / forked feedback topology is likely a dead end for any such attempts, as capacitor coupling introduces up to 90° phase shift, and higher slopes will inevitably introduce unstable regions with positive feedback.
I eventually gave up on the idea, although something might be feasible that resembles linear phase filters with lots of zeroes, i.e.: faking it with notch filters in the stop-band. For the moment I’m leaving it alone. So, feel free to surpass my efforts here.
As proof that the rabbit hole knows no bounds, shortly after writing the above, I discovered yet another possible avenue for development: a balanced bridge design where a small DC current is passed through the speaker. But at this stage I just want to hurry up and build a good-sounding amplifier, rather than dabbling in more theory and simulations for another year.
Transconductance or Current Sensing?
My first amplifier from a few years ago was actually a mixed-mode design. A bit too simple though: open-drain MOSFET output with CCS, and a simple long-tailed pair on the input side. (Too simple because it broke from electrostatic discharge.)
Despite being configured as open-drain, which would have controlled the output current on its own, I used both voltage feedback and current feedback, with the relative proportions varying with frequency. Typical “mixed mode”.
I can foresee two potential problems with current sensing:
- the speaker is placed inside the negative feedback loop. I’m iffy about doing that because of RFI. Maybe it’s a non-issue if suitable filters are deployed, but I hate discovering that the PCBs I ordered are broken and need extensive mods.
- The sense resistor itself is in series with the speaker, so apart from sacrificing a bit of the output power, it can be a significant source of noise. Its high current contributes to this, and it divides the usable voltage down to a small fraction of the total output so the signal-to-noise ratio is a lot worse than it could otherwise be.
Transconductance vs distortion
This time round, I’m inclined to try the transconductance approach: let the MOSFET do what it does best. If open-loop distortion is too high (but what’s too much? I’ve devoted a large part of another post to this exact question…), I’m considering a CFP or Sziklai style output stage.
I’m a little bit scared to do a push-pull topology in open-drain mode. Referencing the power FETs to both power and ground seems a bit too messy. I’d almost prefer a classic common-drain push-pull, which are known to have great performance, and go with the current sensing option.
CS (common source) push-pull seems a bit like a Pass Aleph ‘CCS’ that has been plugged into the input and told to lose weight. Except that the Aleph appears to get its source of modulation from the load underneath (or above), so in that sense its a cascode. CD (common drain) push-pull seems surprisingly difficult if you want to vary the signal amplitude between the two halves — it’s usually designed as class-AB, and there’s only so much bias variation that can occur before one side gets cut off.