Author Topic: vocal megaphone effect  (Read 4223 times)

Rob Strand

Re: vocal megaphone effect
« Reply #60 on: December 17, 2020, 07:23:24 PM »
Quote
I assume I will have to experiment with slopes, stopper resistors to make the range of Resonance and Freq friendly.
The RES CV controls the resonance.    If you can get 2dB peak near cut-off then it's roughly two second order stages cascaded each with Q=1.   Backing off the peak a bit will get you to Butterworth like responses but if you back-off too far it will end-up with a sloppy roll-off.    (It's possible to calculate the values but the datasheets for those special chips take some effect to decipher the finer point.  You can do a spice sim but you still have to decipher the datasheet so the sim matches the chip.)

You will want to play with the resonance anyway but it's good to have an idea where the Q's are set.
The internet:  answers without the need for understanding.

ElectricDruid

Re: vocal megaphone effect
« Reply #61 on: December 17, 2020, 07:53:46 PM »
I assume I will have to experiment with slopes, stopper resistors to make the range of Resonance and Freq friendly.

Yes, you'll have to at least think about it. The "raw" Freq CV response at the pin is 18mV/oct (60mV/decade in the datasheet) so the input divider of 100K/1K8 is designed to drop a much larger CV down into a realistic range for the chip. In practice you need maybe 150mV-200mV at the chip (that's from 8 up to 11 octaves).

The Resonance "CV" isn't actually a voltage at all but a current. The maximum input is 100uA which is supposed to be oscillation, but in practice a bit less is plenty. The Sequential Pro-One uses 75uA maximum, and I rarely use the top third of the Resonance knob's range, so you can trim this to taste.
Use a series resistor (again 100K in the datasheet example) that gives your chosen maximum current somewhere between 50uA and 100uA with whatever your maximum CV input can be.

HTH,
Tom


Rob Strand

Re: vocal megaphone effect
« Reply #62 on: December 17, 2020, 10:49:15 PM »
Quote
The Resonance "CV" isn't actually a voltage at all but a current. The maximum input is 100uA which is supposed to be oscillation,

Yes, fundamentally it's a current controlled input, see fig 6,  but adding the external resistor (100k) makes it a control voltage, see fig 2.  Adding the resistor ensure the control current is limited. 

http://www.bustedgear.com/images/datasheets/CEM3320.pdf

If 100uA is oscillation, probably also with R_RI = 51k as shown in figure 2, it might be possible to back-engineer the Q.   Figure 6 shows gm = 1mS at 100uA control current.   The gm for the feedback implies current out and voltage in for the feedback block; we are talking signal current & voltage not control current now.   From what see R_RI =51k forms a divider with 3.6k ; last column on page 4.     So the signal output current from the resonance block is Ires =  gm * 3.6k / (3.6k + 51k)  Vout = gm / 15.2.   Column 1 of page 4 talks about oscillations and gives a formula for calculating R_RI.
 
The internet:  answers without the need for understanding.

Rob Strand

Re: vocal megaphone effect
« Reply #63 on: December 18, 2020, 12:13:45 AM »
I tried to do a sim and it becomes clear that way the feedback is done to get a resonance does not produce the kinds of filters we would implement if we used opamps with Q of say 0.707 to 1.0.

The filter is inherently two cascaded second order high filters with Q=0.5.  Or equivalently four first order high-pass filters.
As a result, the underlying filter has a very sloppy roll-off.

When you apply feedback it doesn't boost the Q as such.  It's more of a 4th order effect where a peak is produce due to the phase.  I tried flipping the phase of the feedback and it kills the peak altogether.

I'm not sure I've got the feedback scaling correct in terms of the ICR value but it's not going to change the available response shapes.

What seems to be required is to change the structure of the filters from the basically first order topology to second order.    One structure which comes to mind is a state variable filter but that's changing things quite a bit and will need more opamps.   So the question is can we add some feedback around pairs of stages to enhance the Q.

So here's the response I get with no feedback and feedback,





This might be helpful,
http://wseas.us/e-library/conferences/crete2001/papers/327.pdf

Better,
http://class.ece.iastate.edu/vlsi2/docs/Linked%20Publications/1985-03-ICADM-RG.pdf


If anyone can see an error in the sim please let me know.
« Last Edit: December 18, 2020, 01:33:46 AM by Rob Strand »
The internet:  answers without the need for understanding.

Rob Strand

Re: vocal megaphone effect
« Reply #64 on: December 18, 2020, 03:27:59 AM »
Quote
Alternatively, the AS3350 chip is a dual SVF filter, so you could have highpsss and lowpass with separate resonance and cutoff controls for each one.

Datasheet:
https://electricdruid.net/datasheets/AS3350.pdf

Ancient application notes on the original CEM:
https://www.electricdruid.net/datasheets/EMEngCEM3350.pdf

This one has the Q control as part of the chip.  Different structure ... yes, SVF which obviously works!
 
The internet:  answers without the need for understanding.

ElectricDruid

Re: vocal megaphone effect
« Reply #65 on: December 18, 2020, 06:13:08 AM »
I tried to do a sim and it becomes clear that way the feedback is done to get a resonance does not produce the kinds of filters we would implement if we used opamps with Q of say 0.707 to 1.0.
No, it's a "Moog like" structure of four single-pole elements with global feedback around the whole lot. Structurally much more like those allpass filters we were discussing in another thread, in fact.

Quote
The filter is inherently two cascaded second order high filters with Q=0.5.  Or equivalently four first order high-pass filters.
As a result, the underlying filter has a very sloppy roll-off.
It's four first order highpass filters. I don't see anything second order about it, although that may be equivalent.
I don't agree about "sloppy rolloff". Your graph shows a drop of around 75dB in the decade from 1KHz down to 100Hz. That's pretty close to the theoretical 24dB/oct that we expect, and if you measure it better than me eyeballing it, you might find that's exactly what it is.

Quote
When you apply feedback it doesn't boost the Q as such.  It's more of a 4th order effect where a peak is produce due to the phase.  I tried flipping the phase of the feedback and it kills the peak altogether.
I don't understand this. Boosting the feedback boosts the Q, and the Q makes a peak. What did you expect the Q to do if not this?
Flipping the phase of the feedback is obviously going to break things.

Quote
What seems to be required is to change the structure of the filters from the basically first order topology to second order.    One structure which comes to mind is a state variable filter but that's changing things quite a bit and will need more opamps.   So the question is can we add some feedback around pairs of stages to enhance the Q.
*Why* do we want to do this? What's wrong with the perfectly good 4-pole highpass filter you've already got?

Also, if you were to try building a SVF with the 3320, you don't necessarily need more op-amps. Have a look at the Oberheim OB-Xa example here:

https://electricdruid.net/cem3320-filter-designs/


Rob Strand

Re: vocal megaphone effect
« Reply #66 on: December 18, 2020, 08:50:18 AM »
Quote
No, it's a "Moog like" structure of four single-pole elements with global feedback around the whole lot. Structurally much more like those allpass filters we were discussing in another thread, in fact.
Yep, I did see the similarity.

Quote
It's four first order highpass filters. I don't see anything second order about it, although that may be equivalent.
It's pretty common in filter language to bundle two first orders and consider these a second order with Q=0.5.  It then becomes
fairly obvious that the peakyness goes up and the (initial) roll-off goes up as you increase the Q.      Second order Q=0.5 is *identical* to two first orders in cascade which have the same frequency w0.


(this is drawn with the same asymptotic roll-off vs frequency.)

Quote
I don't agree about "sloppy rolloff". Your graph shows a drop of around 75dB in the decade from 1KHz down to 100Hz. That's pretty close to the theoretical 24dB/oct that we expect, and if you measure it better than me eyeballing it, you might find that's exactly what it is.
Sure, but that's the slope some distance away from the cut-off; the so called asymptotic roll-off which depends only on the order of the filter.    The Q has the most effect near the cut-off.  It's a very strong effect.

Here's four high pass filter responses with matching -3dB points
- 1x second order Butterworth (Q=0.707);  total order 2
- 2x second order Butterworth (Q=0.707) in cascade;  total order 4
- 1x fourth order Butterworth (Q=1.307 and Q=0.541); total order 4
- 2 x second order Q=0.5 (critically damped);    total order 4



As you can see the roll-off rate near cut-off of the 4th order 2xQ=0.5 is in the ball-park of the 2nd order Butterworth.    The true 4th order Butterworth has the highest roll-off rate near the cut-off of all the examples.

Quote
*Why* do we want to do this? What's wrong with the perfectly good 4-pole highpass filter you've already got?
The reason why we care is the OP already has a design which is a 2nd order Butterworth.    So going to a 4th order 2xQ=0.5 isn't going to add a lot but it's a lot more complex.

The ultimate aim is to model the horn speaker of the megaphone.  Some horns have fairly steep initial roll-off so the next step up from the existing 2nd order Butterworth design would be a 4th order filter based on highish  Q's

Here's a horn tweeter response.  Obviously the cut-off frequency is higher than a megaphone but look at the roll-off.  About 26dB in an *octave*; so close to the 4th order Butterworth case.


Quote
I don't understand this. Boosting the feedback boosts the Q, and the Q makes a peak. What did you expect the Q to do if not this?
Flipping the phase of the feedback is obviously going to break things.
All I was saying is the way I've drawn it is the best peak you are going to get.    I *tried* the opposite peak in case I'd got the phase of the feedback wrong in the sim.   Also since the loop is 4th order the flip might put a peak in another place.
« Last Edit: December 18, 2020, 11:49:54 PM by Rob Strand »
The internet:  answers without the need for understanding.

ElectricDruid

Re: vocal megaphone effect
« Reply #67 on: December 18, 2020, 01:16:09 PM »
Ok, thanks for the clarifications, Rob. That's helpful.

TBH, I think the 3350 dual SVF is the chip for this job if we're going to more specific chips. You could build two filters with the 3320, but you only have one CV, so they'll necessarily track each other (although that might be quite cool for some purposes, even if not this job). The 3350 frees you from that limitation and gives you more options while keeping it all one the one chip.

Rob Strand

Re: vocal megaphone effect
« Reply #68 on: December 18, 2020, 06:21:07 PM »
Quote
TBH, I think the 3350 dual SVF is the chip for this job if we're going to more specific chips
It would slot into the problem a lot more readily.   I was thinking is it possible to tweak the 3320 circuit to bump the Q.
The internet:  answers without the need for understanding.

Rob Strand

Re: vocal megaphone effect
« Reply #69 on: December 18, 2020, 10:07:49 PM »
OK I've work out a Q enhancement scheme for the CEM3320.   It was derived from scratch but is inspired the decoupled Sallen and Key circuit (which has two buffers to separate the RC sections).

The Q is constant regardless of the frequency setting.

Three examples of the HPF version:
- the out of the box Q=0.5 design
- 2nd order Butterworth  Q=0.707
- 2nd order Q=1.0

I've shown one second order stage.    The idea is applied in pairs of stages.   To get a 4th order response you would use two such schemes.   The w0 and Q of each second order section do not need to be the same for the second order stages.  The Q is set by the cap ratio and the frequency is set by the cap scaling.






EDIT:

This might help understand how I got the filter structure,


« Last Edit: December 18, 2020, 11:01:17 PM by Rob Strand »
The internet:  answers without the need for understanding.

Rob Strand

Re: vocal megaphone effect
« Reply #70 on: December 19, 2020, 02:13:02 AM »
Quote
Quote

    I don't understand this. Boosting the feedback boosts the Q, and the Q makes a peak. What did you expect the Q to do if not this?
    Flipping the phase of the feedback is obviously going to break things.

All I was saying is the way I've drawn it is the best peak you are going to get.    I *tried* the opposite peak in case I'd got the phase of the feedback wrong in the sim.   Also since the loop is 4th order the flip might put a peak in another place.

After getting the new Q enhancement circuit going it started to bug me why my earlier CEM3320 simulations with the standard connections showed those small peaks when using the on-chip feedback.  The slow roll-off with no feedback is to be expected.

I re-did the sim with feedback of both signs but this time with smaller steps in the feedback level.  When I use mild amounts of feedback and with the feedback sign inverted to what I have in Reply #63 I'm getting much more usable looking responses.  To me it's not so clear what the sign of the feedback is on the chip from the datasheet.  Provided I keep feedback control current to 50uA to 60uA max I'm seeing quite a nice response with the standard circuit.

So here's the response the standard circuit but with inverted overall feedback in the simulation and smaller feedback control current settings.


Using the height of the peak for both stages as a measure of the "equivalent" filter Q's:
- ICR = 9.5uA   approx 2x 2nd order Butterworth  (just on the edge of no peak)
- ICR = 25uA    approx 2x 2nd order Q=1.0  (approx 2.5dB peak for both)

I won't get a chance today but it would be nice to compare the overall feedback and enhance Q response shapes when the response peak heights are the same.

So what needs to be found is something in the datasheet which indicates the feedback has negative phase.   The set-up in this post looks too nice to be wrong.
« Last Edit: December 19, 2020, 03:05:21 AM by Rob Strand »
The internet:  answers without the need for understanding.

idy

Re: vocal megaphone effect
« Reply #71 on: December 19, 2020, 01:17:43 PM »
Me reading this back and forth is like a dog watching a ping pong match.

I did pick up that the "CV" inputs are really current, that VCAs have "zero-is" impedance inputs. The app note circuits have 100k series resistors.

Megaphone is one shorthand for the effect. The steep filters(?) on records to give the "radio" sound is another. Both ideas suggest tinny speakers. The beginning of "wish you were here" imitating a radio that then "transforms" into a present sound.

A big motivation for jumping from the 2nd order (with a passive high pass tone control after, three poles, sort of?) to the 3320 is to be able to tune it easily and have something flexible.

I skipped the step of building cascading two 2nd orders. I should try it while I'm waiting on the chips.


Rob Strand

Re: vocal megaphone effect
« Reply #72 on: December 19, 2020, 05:39:10 PM »
Quote
Megaphone is one shorthand for the effect. The steep filters(?) on records to give the "radio" sound is another. Both ideas suggest tinny speakers. The beginning of "wish you were here" imitating a radio that then "transforms" into a present sound.

A big motivation for jumping from the 2nd order (with a passive high pass tone control after, three poles, sort of?) to the 3320 is to be able to tune it easily and have something flexible.

I skipped the step of building cascading two 2nd orders. I should try it while I'm waiting on the chips.

I guess there are tiny and tinny speakers.   The small sound comes largely from the high cut-off.   You should be able to gauge that from your second order box.  The 4th order high-pass filter will help it sound smaller.   The low-pass takes out some of the mess making so you can hear the thing you want to year.

You could experiment the 4dB to 6dB treble (diffraction) boost step I mentioned way back in the thread. This is real effect and is present when you listen things with a physically small size.     Tweaking the boost frequency will let you tweak the "size".    It might help or it might turn out you can bump the high-pass cut-off up a bit to get a similar sound.

The adjustability of the CEM3320 is definitely a advantage when you get to 4th order filters.    Based on the last revision of my simulations with feedback it looks like it should do the job.

I read over the datasheet and I can't confirm the phase of the feedback or add much more to the sim.   Any improvements would only be possible by confirming the sim against the real chip.    The guy that did that chip knew what he was doing so I have faith he's picked the best possible configuration.
The internet:  answers without the need for understanding.

Rob Strand

Re: vocal megaphone effect
« Reply #73 on: December 19, 2020, 07:20:42 PM »
Here's a comparison of the response from the standard CEM3320 configuration with feedback (reply #70) and the Enhanced-Q filters in Reply #69.

The Enhanced-Q filter is two second order HPFs with Q=1.0 in cascade.   The peak of both filters totals +2.5dB.

The feedback of the standard CEM3320 circuit with feedback was adjusted to get a +2.5dB peak.   The caps have the same values as the enhance Q case so the responses of the two filters meet at low frequencies.

The conclusion is the CEM3320 with feedback gives a response peak but it does not produce the increased roll-off slope you see with higher Q filters (and most common high order filters).    The slope near cut-off is very much like 2xQ=0.5 filters. The peak of the CEM3320 with feedback occurs at a frequency somewhat above the Enhanced-Q case about 1.83 times higher.

For comparison, I scaled up the caps of the CEM3320 with feedback case to shift the peak down to the same frequency as the Enhance Q-case.




So I guess the bottom line is maybe the CEM3320 with feedback case doesn't quite get where we want to be for the megaphone filter.

One option is to add Q-enhancement to one CEM stage (fixed to say Q=0.707 or Q=1.0), then use the built-in CEM3320 feedback either around the whole loop or around  the pair of stages (another enhance-Q form but adjustable).    The idea of using the CEM3320 feedback is so the peak height can still be adjusted.

EDIT:
A quick play with the two options in the last paragraph shows the overall feedback doesn't give us the increase roll-off slope near cut-off.   The enhanced Q-form is clearly better.  So basically have one pair for stages wired as an Enhanced Q filter with fixed Q like Q=0.707 or Q=1.0;  that will require an extra inverter opamp.    The next stage we wire the CEM3320 feedback around only a pair of stages (this one is actually the first pair as the CEM3320 feedback is hard-wired there) this stage with have the variable resonance/peak height. Since it looking like the CEM3320 feedback inverts there is no need to add the extra inverter for this pair of stages.
   
Something like the top section of this looks OK,



Adjustable feedback allows peaky and flat responses.

Reference waveform is two second order Q=1.0 filters in cascade (peak approx 2.5dB).
 
« Last Edit: December 19, 2020, 08:41:23 PM by Rob Strand »
The internet:  answers without the need for understanding.

amptramp

Re: vocal megaphone effect
« Reply #74 on: December 20, 2020, 08:55:57 AM »
So far, I haven't seen anyone in this thread suggest regular op amp based bi-quad or other state variable filters.  If you use a multiplying DAC as a feedback resistor, you can tune the filter digitally.  You could put two of them in series to get a fourth-order response but I am not sure that is needed.  This would do the job of the CEM devices without needing single-source parts.

garcho

Re: vocal megaphone effect
« Reply #75 on: December 20, 2020, 12:23:14 PM »
A few random thoughts on a rainy morning:

Maybe a part of the megaphone effect is what the electret mic does to the voice, not just the amp and speaker? Did we already mention that?

I was looking at this thread thinking "it seems kinda familiar... oh yeah, that's right, duh" I used to play in a band where a few of us used megaphones bolted to helmets and played through Smokeys. While I can say it was extremely fun and a big hit with the audience (we would start gigs behind the audience and march up to the stage through the crowd), the sound of an instrument going through a megaphone isn't such a dramatic change in timbre as a human voice going through a megaphone. The accordion and violin had the most dramatic change, but it mostly sounded like distortion and high pass. My electric mando (basically a tiny guitar) and the electric guitar sounded like good old distortion with the "tone" knob full tilt to high. Eventually I made a better amp for the megaphone to be louder and less distorted but it only worked for the magnetic pickup instruments, otherwise it just howled in feedback. We played with a bunch of drummers and brass, so we had real amps and stuff on stage, but our mobile schtick was the megaphone thing. Here's a picture of the speaker helmet subset of Mucca Pazza, when we recorded to wax cylinder at the Edison Laboratory in Orange NJ:


  • SUPPORTER
"...and weird on top!"

ElectricDruid

Re: vocal megaphone effect
« Reply #76 on: December 20, 2020, 05:01:16 PM »
So what needs to be found is something in the datasheet which indicates the feedback has negative phase.   The set-up in this post looks too nice to be wrong.

What I've discovered by playing with the chip is that the filter stages themselves are inverting (very useful for pole-mixing filters) and also that the resonance VCA is inverting. So the resonance feedback path *is* inverting, because the resonance VCA flips it.
I found this because I was experimenting with doing resonance volume-drop compensation (In many synth filters as resonance increases, passband gain decreases). The typical way this is done is to feed some input signal to the resonance VCA as well as the input, so that as the resonance increases, the input level also increases to compensate the passband drop. I found I needed to use a differential mixer to combine the output signal and the inverted input signal. This implies the resonance VCA is inverting, putting the signal back in phase with the main input. More details here:

https://electricdruid.net/wp-content/uploads/2020/05/3320MultimodePg1-scaled.jpg

From https://electricdruid.net/multimode-filters-part-2-pole-mixing-filters/

ElectricDruid

Re: vocal megaphone effect
« Reply #77 on: December 20, 2020, 05:02:44 PM »
BTW Rob, I love the Sallen-Key sections done with the 3320. That's excellent work, nice one. I've not seen that done before.

 8)

garcho

Re: vocal megaphone effect
« Reply #78 on: December 20, 2020, 10:00:44 PM »
That resonance compensation mixer is brilliant, i have a 3320 on the breadboard, it works like a charm, very welcome addition to the circuit.
  • SUPPORTER
"...and weird on top!"

ElectricDruid

Re: vocal megaphone effect
« Reply #79 on: December 21, 2020, 07:54:00 AM »
That resonance compensation mixer is brilliant, i have a 3320 on the breadboard, it works like a charm, very welcome addition to the circuit.

Thanks!