Stereo hypertriangle LFO progress

[MrStab]

Hi guys,

This is gonna be a hard one to go over. Bear with me.

First off, a summary: Thanks to a previous post, some serious flaws were pointed out in my original idea for inverting the Ross Phaser OTA LFO for stereo use, so i started from scratch using different building blocks. what i needed were two hyper-triangles/pseudo-sinusoidal/drawings of mountains a half-cycle apart in either channel; as opposed to peak in one channel, trough in the other.

The basis for the circuit i'm currently working on is the Boss CE2 LFO: http://www.electrosmash.com/images/tech/ce-2/boss-ce-2-lfo-circuit.png and the latter half of this method of de-linearising the output: http://i.imgur.com/OVFTOYq.png . Through messing about in LTSpice, which i'm very new to, my nano-brain realised that using PNPs and switching the inputs in the hyper-triangle section of the circuit could result in the out-of-phase waveform i was after.

Here's the circuit so far - apologies for the messy layout, i keep tweaking it:



I do intend to tweak the values a bit more for maximum sweep vs. being reasonable with a 9V battery, but my main issue at the moment is symmetry.

POINTS OF NOTE:

• All 10k resistors are 11k on my breadboard because i ran out, damnit.
• R25 and R26 are a Depth pot, shown at minimum depth. C5 seems essential or the waveform gets far too crazy, though it does still affect the shape at minimum setting (more triangular trough). I can't tell to what extent the benefit is from decoupling or from filtration.
• R5 on the PNP side is 22k as opposed to R9 (15k) on the NPN side, as this seemed to even out the Vpp of the sweep.
• Originally, there was a weird kinda jump in the bottom half of the PNP waveform. So:
• D1 exists to smooth out the PNP waveform by offsetting the diode drop of Q5 & Q6 (i think...). Only R5 returns to V+ pre-diode. Still not perfect, but much less of a sudden jump.
• C3 & C4 in the feedback loops of U3 & U4 exist to smooth this out further.
• I'm actually using BC549C's in the NPN half instead of 2n3904's, though i intend to replace em.

on the whole, conceptually, this thing works great. the only problem i'm having is symmetry, by which i mean L channel will peak, R will peak, then there's too much of a gap before L peaks again. previous iterations with different values (notably uneven Collector values) seem to be better, but the waveform in LTSpice just doesn't display as severe a mismatch as what i'm hearing, so it's hard to pinpoint the exact components. i realise the rise and fall of each wave isn't entirely symmetrical to begin with.

so my only question really is this: what should i tweak to fix the symmetry, if possible? as this is kinda "misusing" differential amps for a certain effect, i'm finding it difficult to find literature i can apply to this context. could it be the BC549's i'm currently using, which have over double the gain of the 2n390* family? or those Collector values? HFE has been matched between pairs, at least. i'll upload a rough clip of the sweep in either channel later - maybe it's actually fine and i'm talking crap. i have no scope, which is an issue.

also, any comments, ideas... go for it! and feel free to adapt & use this thingy, if any of you guys have a use for it. i'm standing on the shoulders of giants anyway. FWIW, it still sounds totally awesome to play with headphones and small amps, mismatch or no.

(off-topic, but JFET-controlled attenuators work surprisingly well for controlling 2 regen channels with 1 pot, btw!)

[MrStab]

Clips:

Some nice ground buzz to show the sweep generally:
https://www.dropbox.com/s/80bh1w6d8vq1rr4/sweepmismatch1.mp3?dl=0

Simple riffs to show the sweep on a guitar signal:
https://www.dropbox.com/s/kdsjdikjxiliw8i/sweepmismatch2.mp3?dl=0

[MrStab]

lol this is my 2nd thread on this subject where i'm just talking to myself like a nutjob. good thing i am actually a nutjob. sorry for the spam, i guess i'm just a bit excited as i usually make circuits that don't wobble.

discovery: forget the PNP section for a second. duplicate the NPN section, but in the copy, move that 15k resistor (R9) to the other Emitter (Q1) and switch the opamp inputs. the waveform/polarity is the same but shifted by 180 degrees, minus the weird irregularity in the trough of the PNP version. the input resistor (R8) and biasing needs to be increased for two NPN stages to work in parallel, seemingly. starting to understand more about this. still dealing with slight asymmetry, seems one of the slopes is always veeery slightly off.

will tweak some more. get the straitjacket out, guys - post #4 is on the way!

Edit: ah, forgot tweaking the bias can fix the symmetry, whose source is the integrator. gonna try and get this method working as it seems preferable to the NPN & PNP method. currently trying to get bottom of sweep to be less flat.

[garcho]

i might be way off suggesting this, but what about using two synced VCLFOs or TAPLFOs, and invert the clock from one to the other? you could fine-tune with wave distortion.

[MrStab]

can't seem to open the links unfortunately, Gary - just a blank page, no 404. maybe the server's down or something?

i'm fairly-comfortable in my understanding of most of the current circuit, plus it seems to work pretty well for the most part, but if i can't even out the channels i'll definitely look into your suggestion. i'm abandoning my awful cheap breadboard, so i just need to wait a coupla days for some stripboard to try the double-NPN-stage thing. maybe wasteful, but i vastly prefer experimenting on it.

a key principle with all this seems to be invert once before and after processing, else the sharp peak will just become the bottom. thanks to PRR for leading me to realise that. not sure why i'm mentioning it, but it feels relevant. lol

cheers! i'll check back and keep trying those links.

[PRR]

Rather than offsetting R9 R18 from R5 R21 (which really should match), maybe what you want is to diddle the R6 R7 ratio to "center" it off-center your desired amount?

This may lead to a different R6 R7 -like divider to set bias on U1 U2; or maybe not.

R4 R25 R26 R27 C5 appear to do nothing with the values shown. If R25 were larger, it would be loaded by the stuff hung off it; but 1 Ohm feeding >100K can't do anything significant.

> LTSpice, which i'm very new to

I don't know it; but my older SPICE has "bubbles". I tie one to my +9V, give it a name like "+9V", and then copy that symbol to places where I want +9V. Saves a lot of long lines running around the diagram.

Not sure what D1 does for you. I'd expect everything to work off the one voltage. Reverse-battery protection may be useful in real life. I'd then want it to protect the Q5 Q6 corner as well (although this part won't burn with reverse battery).

For simulation, make R6 R7 much lower than the stuff they feed. 1K, even 100r. In SPICE you can make them 1 Ohm, because a SPICE battery never dies and does not sag (if Rser is 0). At some later point you need to pick "realistic" values, true. But for SPICE, a solid reference is possible and avoids some unsuspected problems.

[MrStab]

thanks for the reply, Paul. i'm going for a slightly-different approach from my OP, briefly hinted at by my 2nd post, but i think your comments still apply. here's where i'm at - there's some scruffiness such as a voltage divider to nowhere, but the concept is there:



as usual, i've left too much to assumption. as i couldn't get an imported pot model working in LT, R4 is meant to represent a Rate pot at full-rotation (with 10k limiting resistor, hence 110k), and R25, R26 a Depth pot. i plan to alter these values for usability later, as that's charted territory for me whereas the rest of the wave-shaping isn't. the lower the value C5 is, the less-skewed the signal when the depth pot is lowered, so i'm pretty sure there's some filter action going on there.

D1 in the original diagram basically existed as the trough of the PNP-portion of the circuit had this weird and audibly-noticeble bump in the waveform, which i could only attribute to a diode drop. adding D1 seemed to clear it up a lot, but there was still a lop-sided curve. the current approach is far better-matched in either channel.

i'm in a wee bit of a rush at the moment but i'll be back in a few hours to take things on board some more.

cheers!

[MrStab]

actually, as my approach has changed and because i wasn't clear about the grouping of some resistors as pots, i think i've pretty much covered all those bases. except:

I don't know it; but my older SPICE has "bubbles". I tie one to my +9V, give it a name like "+9V", and then copy that symbol to places where I want +9V. Saves a lot of long lines running around the diagram.

yeah, i got lazy with trying to actually find the single voltage port and rushed straight into things. should get that fixed for posting when it's done.

For simulation, make R6 R7 much lower than the stuff they feed. 1K, even 100r. In SPICE you can make them 1 Ohm, because a SPICE battery never dies and does not sag (if Rser is 0). At some later point you need to pick "realistic" values, true. But for SPICE, a solid reference is possible and avoids some unsuspected problems.

duly noted, about the resistor values and also that battery resistance should be 0. wasn't sure if it was worth simulating real-world conditions with the internal resistance, but you make the point that it could be intrusive to testing.

there's a slight different in peak voltage between channels. may be small enough to overlook without attenuation, though. i'm guessing it's a consequence of reversing the U3 & U4 inputs, or where that Emitter-ground resistor goes?

the 470n caps on each output aim to act like Mark Hammer's Ross Phaser mod where a cap to ground is added after the LFO output resistor to round off the sweep a bit at higher speeds. nothing special bout 12k, it was just the first resistor baggie i picked up at the time. just another placeholder for 10k. iirc my calculation for that roll-off (i think Mark said 0.3Hz so i aimed for that) produced 47uF for a 12k resistor, but by adjusting the Rate "pot" and counting cycles over one second in the simulator, it seemed to take 0.47uF to do the job. i've definitely screwed that up and made a dumb mistake, haven't gone back to look admittedly.

unbalancing the bias seems to do the trick for balancing the wave, and the sweep's still far enough off the negative rail even if it doesn't pass right through the middle. i wouldn't mind increasing the swing, but i think the range is pretty decent as i have it now.

to elaborate on C5: it seems to be needed because the response from the transistors seems much more (but not quite) linear when they receive less input, but a low-ish value cap compensates. not entirely sure how rate-dependent this is. there's the consequence of softening the waveform at the peak as the Depth pot (R25/R26) shunts more to Vref, but there's probably some ideal compromise. or better way of doing it. i don't want to use dual-gang pots because this'll be using those Alpha long-pin ones.

hopefully this thing could be pretty cool and usable on a bunch of phasers. OTA & fixed stages for me. my stripboard hasn't arrived so i haven't tested the new configuration for real yet.

cheers!

[MrStab]

rebuilt on stripboard, managed to even out the channels by tweaking the bias a bit. i tweaked it by a tad too much, forgetting i used 11k in lieu of 10k resistors, but still pretty usable. bias ratio is currently 11k by 8.2k.

coupla recordings. crap riffs and i wasnt bothered to find a pick, and the regen is a bit mismatched on either side but i'll get to that:

Guitar riffs: https://www.dropbox.com/s/oyifgrkctwm71hy/phaserprotstereo3a.mp3?dl=0

Sweep noise: (a bit loud! turn down!): https://www.dropbox.com/s/8l1xmgfuvqugoam/phaserprotstereo3b.mp3?dl=0

oh, i also raised R9 and R21 on my previous diagram to 22k to avoid hitting the opamp rails. the sweep is now a measly 3Vp-p, but i can fix that, i just want to make sure the concept is solid. seems to be...

also, i forgot to match the HFE of each transistor and soldered them in, but according to SPICE and my ear, that's not such an issue