Narrowcast V2 [documentation]
Posted: 30 Jun 2021, 20:20
It's been a year and a half since I made the Dudson Narrowcast, a reworked version of the popular Hudson Broadcast. It was fun and sounded nice, but it wasn't perfect, and I left it sitting there. The issue I couldn't gulp down was that i had to omit the feedback lowpass capacitor because of oscillation issues. That and more has been fixed in this V2. This update is also an intriguing perspective in how much I learned since then. There's always more, but at least I feel like I understand the circuit enough to know what to do.
The changes are pretty substantial so that I didn't list them one by one anymore, but in theory the sound should be the same. The first important thing is that as the original, it's a series-shunt feedback amplifier, with AC only global feedback. This is what I rebuilt this around, and as you notice the first stage is greatly simplified. Gone is the bypassed 5.6K emitter resistor, since it doesn't affect AC operation. R5 stays as is, and it's bypassed by C6 in variable amounts depending on the gain mode switch (which also means different amounts of feedback). R9 serves pretty much the same function as the 56 ohm in the original, but it's now in series just with the bypass capacitor because it seems more elegant (makes no difference for DC bias since it's in series with 1K). The base bias is simplified too. Instead of taking DC feedback on a tap on the collector, I've taken it directly at the collector. R3 and R2 have been chosen to give the desired bias, with R3 being large enough to not cause loading, and R2 to keep the original input impedance. This feedback is DC only, so it doesn't affect gain or input impedance. Since the base voltage needs to be low, the two resistors are enough to drop it.
I've removed some degeneration on the emitter, but I actually have more DC feedback from the collector. How does this affect bias stability? It actually doesn't that much! Probably the coolest characteristic of the original is there pretty much intact: bias is almost impervious to supply voltage changes or beta variations, which is why the original can run at 27V too. In simulation, the collector voltages were pretty much multiplied by the same amount as the supply is, with Q2 being 19.5V at 27V supply (6,5*3=19.5!
). In practice this means that you can try any voltage up to 27V and any combination of transistors and you can expect the circuit to behave as intended with no changes.
Q2 is pretty much the same as the original, with the obvious exception of being NPN. I'm not preventing anyone to try this with PNP transistors, so for that there's the version below. The Q1 has been rebiased to have about 6,5V on Q2's collector (why not 4,5V? the original biases at about 2.5V, but being NPN instead, the same bias should be taken from the positive rail, so that gives 6.5 the off-center bias is an important part of the sound so I kept to it), and simulation shows the headroom is the same as the complementary version, so I guess I've done things right.
Some things are gone, some things are back. Like the power supply filter resistor, or the gain stopper that prevents the output from cutting off, or C11. Oh how glad I am to get C11 back. I never had luck and always got oscillation with it, no matter if I was breadboarding the narrow, the broadcast, with silicon or germanium, with or without transformer. In the end some general bandwidth limiting/dominant pole compensation did the trick with C5, which doesn't otherwise affect response much at all. I guess you really need poor bandwidth to begin with otherwise, and even my germs were too good...
Did I say transformer? That one is on the bench still. I even got a proper "10K" transformer and made comparison, and I didn't hear anything so I keep my stance. And since it makes the circuit more affordable and easy to build, I'm all for it. A 25K or 50K volume pot gives a similar load as the original, accounting for all the series and parallel output resistors. Taper is log.
As it's usually the case for me, values have been standardized to the E6 series where this doesn't mean a compromise, with gain and AC response (in this case I've also done transient analysis) as close as the original as possible. Here's a sweep of gain and bass controls comparing narrowcast and broadcast (minimum and maximum bass, four gain settings) in low gain mode:
and high gain:
Here's the PNP version. The change in the Q1 base bias gets reflected in the rest of the circuit and biases Q2 to the desired 2.5V with no other changes:
Here's a compact and verified vero layout:
And finally demo:
I've removed some degeneration on the emitter, but I actually have more DC feedback from the collector. How does this affect bias stability? It actually doesn't that much! Probably the coolest characteristic of the original is there pretty much intact: bias is almost impervious to supply voltage changes or beta variations, which is why the original can run at 27V too. In simulation, the collector voltages were pretty much multiplied by the same amount as the supply is, with Q2 being 19.5V at 27V supply (6,5*3=19.5!
). In practice this means that you can try any voltage up to 27V and any combination of transistors and you can expect the circuit to behave as intended with no changes.Q2 is pretty much the same as the original, with the obvious exception of being NPN. I'm not preventing anyone to try this with PNP transistors, so for that there's the version below. The Q1 has been rebiased to have about 6,5V on Q2's collector (why not 4,5V? the original biases at about 2.5V, but being NPN instead, the same bias should be taken from the positive rail, so that gives 6.5 the off-center bias is an important part of the sound so I kept to it), and simulation shows the headroom is the same as the complementary version, so I guess I've done things right.
Some things are gone, some things are back. Like the power supply filter resistor, or the gain stopper that prevents the output from cutting off, or C11. Oh how glad I am to get C11 back. I never had luck and always got oscillation with it, no matter if I was breadboarding the narrow, the broadcast, with silicon or germanium, with or without transformer. In the end some general bandwidth limiting/dominant pole compensation did the trick with C5, which doesn't otherwise affect response much at all. I guess you really need poor bandwidth to begin with otherwise, and even my germs were too good...
Did I say transformer? That one is on the bench still. I even got a proper "10K" transformer and made comparison, and I didn't hear anything so I keep my stance. And since it makes the circuit more affordable and easy to build, I'm all for it. A 25K or 50K volume pot gives a similar load as the original, accounting for all the series and parallel output resistors. Taper is log.
As it's usually the case for me, values have been standardized to the E6 series where this doesn't mean a compromise, with gain and AC response (in this case I've also done transient analysis) as close as the original as possible. Here's a sweep of gain and bass controls comparing narrowcast and broadcast (minimum and maximum bass, four gain settings) in low gain mode:
