[sdiy] Compensating output level for Q

Fri May 31 15:07:37 CEST 2013

Hi Nicholas,
first off, the email that Richie forwarded hasn't got inline images,
so I'm missing out a bit on what Bernie wrote and therefore can't
comment on all of it.

However, if you have got a 4p HP that has no resonance, try the following:

1. get a noise source and a spectrum analyzer (Voxengo Span will be
great) and see what's happening in 4p HP mode. See if the peak can be
seen at all, expecially if you change the Q.

2. hook the noise input into the first stage, hook the specan to the
output of that first stage, make feedback go from the output of the
first stage to its input, and see if that works, and if it has
resonance. Calculate the frequency peak.

3. hook the noise input into the second stage, hook the specan to the
output of that second stage, make feedback go from the output of the
second stage to its input, see if that works and if it has resonance.
Calculate the frequency peak (make sure you haven't touched the cutoff
knob from when you measured it in step 2).

If you get resonance in 2 and 3 but not in 1 you want to check two things:

1. Is the resonance in both stages, at the same frequency setting,
happening at the same frequency? If not, trim your filters.

2. Is the resonance in-phase? Sometimes the filter will need resonance
out of phase because it goes into the negative input of a summing amp.
Or sometimes it's the other way around. If you have one filter stage
which requires negative resonance input, then the partial at the
output which corresponds to the resonant frequency of the filter
should be out of phase compared to the input. You can check that with
a bode plot. If you have two such filters and you use global feedback,
then what happens is the first stage puts the frequency out of phase,
and the second stage puts it in phase again, so in the end the global
feedback input gets it in phase, which might be incorrect for your
filter. So check that. One way to fix this would be to connect the Q
input to the positive input of IC41a (only when you're doing 4p HP).
Another way would be to use an SVF output for feedback such that the
feedback frequency gets negated an odd number of times (again, only
for 4p HP here). Another way is to use a unity inverter amp in the
feedback amp (4p HP again). All of those are kludges though. I have a
feeling that your SVF changes this in-phase/out-of-phase relationship
of the resonant frequency depending on the mode used. If you look into
that you might be able to work out exactly what you should be doing.
Just check which filter modes flip the resonant frequency. If you
don't have a way to make bode plots (and it's quite simple, there's
loads of software for creating that which I can't name ATM!) then you
can just take a sinewave, and record the output from the filter, and
record the dry sinewave. Tune the sinewave to the filter's resonant
frequency and mix both. You should either see a 2x increase in
amplitude, or it should zero out. If you see something else then the
sinewave is at the wrong frequency. If it nulls out then the filter
mode flips your resonant frequency, if it doesn't null out then your
filter mode does not flip your resonant frequency.

Cheers,
D.

On Fri, May 31, 2013 at 2:11 PM,  <rburnett at richieburnett.co.uk> wrote:
> The plots that Bernie included in the Email you forwarded to me are called
> Root-locus plots.  Root locus is a technique used by control systems
> engineers to determine how the poles of a system move as feedback is applied
> around the system.  This is usually done to get a feel for how stable a
> system is when feedback is applied, and what the bandwidth will be.  It's a
> very common engineering tool, and it's useful for analysing analogue filters
> (and also digital filters!) for musical applications.  Specific types of
> filter like 2-pole SVF or Moog 4-pole cascade exhibit very recognisable
> root-locus shapes like circular arcs or X-shapes, and this is one of the
> reasons why all lowpass filters don't sound the same.
>
> So in order to answer your many questions about why certain configurations
> exhibit resonance and others don't, it's probably time to get out a control
> theory book and brush up on maths like Laplace functions, AC analysis and
> root-locus.  These techniques combined with a Pspice circuit simulation
> package, are very powerful for trying out different filter connections and
> gaining an understanding for why some combinations work, and others don't do
> what you had expected!
>
> -Richie,
>
>
>
> On 2013-05-31 02:00, Nicholas Keller wrote:
>>
>> Richie, Tom, Olivier, and anyone else....
>>
>>  I'm straying from my original question now, but if you're at all
>> interested in some other things I'm trying to squeeze out of this
>> filter, I'd be grateful for your, well, feedback.  I've been in touch
>> with Bernie Hutchins who seemed very interested but veered a bit
>> off-topic before losing touch (I'll paste my conversation with Bernie
>> below, you might understand it better than I), and more recently with
>> Carmine Bonanno, the designer and founder of
>> Octave-Plateau/Voyetra/Turtle Beach, but he really doesn't recall the
>> circuits off-hand and has no archive, though he said he was willing to
>> field some questions.  I just haven't had time to ask him.
>>
>>
>>
>> Being that it is two cascading SVFs, I'm trying to make it
>> multimode.  The Cat SRM was originally offered with a series of
>> optional factory mods.  You can see the list of mods here:
>>
>> http://niroke.blogspot.com/2013/03/octave-cat-srm-factory-modifications.html
>> [1]   Mine came with the 12/24dB mod and the HP/LP mod, however these
>> did not work properly.  I managed to figure out why the 4pole HP mode
>> didn't work.... because the mods didn't switch the mode from the first
>> section going into the second, so the 2pole LP always fed the second
>> stage which made the 4pole HP disappear.
>>
>>
>>
>> Maybe I should have left it at that, but when I found out I could
>> also tap BP modes, and perhaps a Notch, I wanted to max it out.  I set
>> up a 2pole, 6-way rotary switch to select the mode feeding the second
>> stage and the tap point feeding the VCA. It looks like this:
>> http://i.imgur.com/gtyGktQ.png [2]
>>
>> The problem with this configuration is that the BP mode is not as
>> loud as the HP and LP modes, some modes are inverted and some aren't,
>> but mainly the 4pole HP mode has no resonance at all.  I'm thinking
>> this might have something to do with the fact that the feedback is
>> from the 4pole LP tap, but in HP4 the second stage is fed by HP2.  In
>> this case LP4 has little to no signal.
>> http://www.youtube.com/watch?v=p_WkSj14C-k [3] Here's a video showing
>> that the modes do work, though it's hard to tell a difference between
>> any of the 12 and 24dB modes, except that HP4 has no resonance. That
>> is at the very end and is only for a few seconds.  The Notch I tried
>> didn't have much effect, Bernie says that accurate components are
>> important to achieve this mode, and I have purchased 1% resistors and
>> 5% caps for the filter section but haven't installed them yet.
>>
>>   I invested in some Electronotes from Bernie, and a couple of his
>> "books", and noticed a few variations of SVFs and several filters
>> based on the SSM2040.  In some of the filters, the feedback was tapped
>> from the BP4 point.  Can anyone explain the benefit of this, or the
>> difference it would make.  Is this the necessary feedback source if
>> wanting to have modes other than LP?  Other versions of the Dual SVF
>> design uses a separate feedback path for each section.  I tried
>> several different ways of configuring my SRM board (feedback from LP2,
>> BP4, HP4(for HP4 mode only) and also sending the selected mode output
>> to the final inverting amp in the filter circuit rather than to the
>> VCA) but got poor results.  Either there was no resonance (presumably
>> because the point being tapped was inverted from the input, this
>> wouldn't work right? Although, Richie, in your reply you mention
>> global negative feedback) .... there was a strange cross-modulated
>> effect occurring like two FM'd oscillators slowing to a stop .... or
>> with self-oscillation occurring, adjusting the cutoff above or below
>> (I don't remember now) a certain point would cause the feedback to
>> disappear until Q and Fc were reset to medium levels.  I should have
>> better documented these results with video, etc.  I can if you think
>> it will be helpful.
>>
>> I'd really appreciate any help you guys can give me, suggestions of
>> things to try, etc.
>>
>> See below for my email to Bernie and Bernie's thoughts, though some
>> of it is repeated from above,
>>
>> nick
>>
>> ------------------------------
>>
>> Hello, Mr. Hutchins.
>>
>>  I recently received my box from you containing the MEH and PCC, as
>> well as the Electronotes you used to fill the box. Some of those pages
>> seem to have specific relevance to a project I am working on. I had
>> already done some experimenting before I ordered this literature from
>> you, and had seemed to be stumped. Now I see that there may be good
>> reason that my circuit is not completely functional. (ELECTRONOTES
>> VOLUME 10, NUMBER 85, PAGES 12-17:  "OBTAINING BANDPASS AND HIGHPASS
>> FUNCTIONS FROM A FOUR-POLE LOWPASS" AND "PRESENT DAY VCF DESIGN
>> OPTIONS",  JANUARY 1978)
>>
>> http://i.imgur.com/gtyGktQ.png [2]
>>
>> The link above is an edited version of the OCTAVE CAT SRM FILTER
>> board, with an added 2x6 rotary switch. I bought this keyboard with
>> factory modifications installed, however, they were not performed
>> properly. The modifications included a switch to select 2 or 4 pole
>> slope, and a switch to select HP or LP mode, so that's four mode in
>> all. The 4poleHP mode had almost no output. I began trying to work out
>> how the mod should have been done.
>>
>> Someone told me that the SRM's 2040-based filter was a state variable
>> design, and that 2 and 4pole versions of HP, BP, and LP were available
>> to be tapped. The locations on the image above are where he told me to
>> tap them, however, he drew me an image with a single pole, 6-way
>> rotary. This didn't make sense to me, I thought that the input to the
>> second 2pole filter stage needed to be switched to match the selected
>> 4pole output. This is why the factory 4poleHP mod didn't work, the
>> input was always LP. The image above shows how I wired my 2x6 rotary
>> as well as where I cut traces to return my switch outputs. I did
>> attempt to tap LP4 and return the selected filter mode to the junction
>> of SSM pin10/R175 and R183, after cutting the trace there of course.
>> But that didn't seem to work either.
>>
>> All the modes appear to work, except that THE 4POLEHP MODE HAS NO
>> RESONANCE. Also, when I attempted to create a Notch mode, it had
>> practically no effect. I noticed that you say the tolerance of
>> components are important to achieve a Notch mode, otherwise it becomes
>> "washed out" (page 15). I am planning to replace all the resistors
>> with 1%, and will attempt to replace the capacitors with more accurate
>> caps. I don't know that this will solve my problem with the HP4 Q. I
>> notice in some of the other state variable filters in Electronotes
>> (like inEN#84, PAGES 17-18  "READER'S CIRCUITS - VCF WITH SSM2040 BY
>> RICH RYAN), the feedback is drawn from the Bandpass output, rather
>> than the Lowpass output. The person who told me about the tap points
>> said that the feedback HAD TO come from the Lowpass output. This
>> confused me as well, because if my rotary is switching the input to
>> stage 2, when I select HP4 mode (therefore HP2 feeds stage 2), there
>> would be very little signal at the LP output to create resonance (much
>> like the non-working factory mod). I tried to use the BP4 output for a
>> feedback source, but that didn't provide the HP4 mode with resonance
>> either. The BP modes don't seem to have the same volume as the LP and
>> HP modes. In FIGURE 21 ON PAGE 17, and on PAGE 16, it is stated that a
>> dual state variable REQUIRES SEPARATE RESONANCE CONTROLS FOR EACH
>> SECTION.   Could this perhaps be part of my problem with HP4's Q?
>>
>> -----------------------
>>
>> Hi Nick -
>>
>> Good questions. Let me try to begin answering it with a few comments
>> and we can go from there. I don't understand your questions
>> completely, which reflects not on your ability to ask a question, or
>> so much on my inability to put my mind in gear, but the fact that the
>> matter is complex and better attacked in parts.
>>
>> First of all, the diagram you sent is correctly thought out. You are
>> correct that you need the two-pole six-position switch. Putting a
>> low-pass and high-pass in series would be a no-pass. Thus in the
>> correct version you sent you are suggesting a classic cascade approach
>> to fourth-order. The following comments may be useful.
>>
>> (1) For the moment, we assume that both state-variable sections have
>> the same frequency and the same Q. They are identical. Later you can
>> change these with additional possibilities (probably too many
>> possibilities).
>>
>> (2) I am not sure a fourth-order bandpass is worth the trouble. You
>> are usually using bandpass with high Q, and whether the "skirts" are
>> 6db or 12db probably matters little. If you were, however, to offset
>> the two filters slightly in frequency (which we are agreeing to
>> postpone), you could toy with the idea of a bandpass plateau rather
>> than a single peak.
>>
>> (3) With low-pass cascading to low-pass, you indeed get a
>> fourth-order low-pass, but NOT a Moog low-pass in general. You simply
>> get the product (square in this case) of the second-order low-pass.
>> You could set the Q's to different values, and get, for example,
>> Butterworth, but there is nothing especially useful about Butterworth.
>> So this cascade is likely sharper than Moog, two pairs of higher Q
>> poles instead of Moog's single pair (heading for the j-omega axis)
>> with a second pair moving away (on the corners of a square - see first
>> fig). The second pair of poles (facing away) result in the eventual 24
>> db roll-off. I suspect that the "mystic" values of the Moog is due to
>> this delayed roll-off. (It could also be imaginary!).
>>
>>  figure shown for gains of 0, 1/2, 1, 1.5, 2, 2.5, 3, 3.5, and
>> finally 4. Different shapes and colors just to separate poles.
>>
>> (4) You CAN achieve a classic Moog (as fig above) exactly with your
>> setup, if you implement the feedback loop. I am not sure if you say
>> you tried this and failed, or not? If you set the Q of the two
>> sections both to 1/2, you have four first-order low-pass sections in
>> series, and the feedback around it peaks and eventually oscillates at
>> a loop gain of 4. Of course, the polarity of the feedback needs to be
>> right.
>>
>> (5) So far, this is not new stuff. But out notion of using two-state
>> variables with Q=1/2 was just to get the classic Moog 4-pole back. I
>> never tried it any other way, but this is a good chance. It only took
>> me 15 minutes to write the code for the figure above, and then you
>> just change one number. Suppose the original Q's were Butterworth,
>> Q=l/sqrt(2). We get the next figure:
>>
>> Here there are two pair of poles starting at -1/sqrt(2) +/- j
>> /sqrt(2) when the gain is 0,(instead of four poles at -1). As gain
>> increases to 1/2, 1, 1/5, and then to 2, the poles move to the j-omega
>> axis at a frequency of 1. I didn't know this until I just now tried
>> it! Thanks. So, as one might expect, if we start with a higher Q, we
>> get oscillation at a lower gain (2 instead of 4). Note that
>>
>> 2 * (1/sqrt(2))^2 =1. Neat.
>>
>> So first thanks for prompting me to look at this. I may well add it
>> to some future report. In the mean time, I probably have told you some
>> things without answering your actual question(s), so please DO keep in
>> touch.
>>
>> Bernie
>>
>>
>> -----Original Message-----
>> From: Richie Burnett [mailto:rburnett at richieburnett.co.uk]
>> Sent: Thursday, May 30, 2013 6:42 PM
>> To: Tom Wiltshire; Nicholas Keller
>> Cc: synth-diy at dropmix.xs4all.nl
>> Subject: Re: [sdiy] Compensating output level for Q
>>
>> Yes, two cascaded 2nd-order state variable filters to give an overall
>> 4 pole response. Then with global feedback around it to achieve
>> resonance like in the classic moog ladder.
>>
>> Any interesting behavior might come about if the two state variable
>> filters are designed with different Q factors. But I think any
>> difference would probably become quite small once the resonance pot is
>> increased and global negative feedback is applied around the SVFs.
>>
>> -Richie,
>>
>> Sent from my Sony Ericsson Xperia ray
>>
>> Tom Wiltshire <tom at electricdruid.net> wrote:
>>
>>> Interesting filter design, that. What is it exactly? Comparing it with
>>> the SSM2040 data sheet, it looks most like a couple of 2-pole SVFs stuck end
>>> to end, with feedback round the whole lot - a bit like the Jupiter 6. Or am
>>> I completely wrong?
>>
>>
>>>
>>
>>> Tom
>>
>>
>>>
>>
>>> On 30 May 2013, at 22:27, Nicholas Keller <niroke at tampabay.rr.com> wrote:
>>
>>
>>>
>>
>>>>
>>
>>>> Richie,
>>
>>
>>>>
>>
>>>>   It's the deafening squeal one.  The synth is an Octave Cat SRM with an
>>>> SSM2040 based filter.  Here is the schematic:
>>
>>
>>>>
>>
>>>> http://i.imgur.com/R028I6l.jpg [4]
>>
>>
>>>>
>>
>>>> I see D30 and D31 in the SH-101 schematics, so I could try that.  Looks
>>>> easy enough.  I don't know what an equivalent diode is to that: 1s2473.
>>>> I'll look it up online if you don't have a suggestion.  The 101 has other
>>>> parts between the Q wiper and the diodes though, TR26 and TR27, etc.  I
>>>> don't know what those do....
>>>> http://manuals.fdiskc.com/flat/Roland%20SH101%20Service%20Manual.pdf [5]
>>>> page 9
>>
>>
>>>>
>>
>>>> If you hadn't guessed, I'm rather uneducated in regards to electronics.
>>>> I've built a few things, modded a few things, but I still don't "get it".
>>
>>
>>>> I appreciate your help.
>>
>>
>>>>
>>
>>>> nick
>>
>>
>>>>
>>
>>>>
>>
>>>>
>>
>>>>
>>
>>>>
>>
>>>> -----Original Message-----
>>
>>
>>>> From: Richie Burnett [mailto:rburnett at richieburnett.co.uk]
>>
>>
>>>> Sent: Thursday, May 30, 2013 4:39 PM
>>
>>
>>>> To: Nicholas Keller; synth-diy at dropmix.xs4all.nl
>>
>>
>>>> Subject: Re: [sdiy] Compensating output level for Q
>>
>>
>>>>
>>
>>>> Do you want to compensate for the perceived drop in overall volume as
>>>> the resonance is increased, or compensate for the sudden increase in volume
>>>> at the instant where the filter commences self-oscillation?
>>
>>
>>>>
>>
>>>> The former is often compensated by feeding a fraction of the input
>>>> signal into the feedback path around the filter, before the resonance
>>>> control pot. Essentially this acts to increase the filter drive level as the
>>>> resonance control is increased and the pass-band gain of the filter
>>>> naturally decreases. You can compensate some fraction or all of the volume
>>>> drop depending on how much of the input signal you feed into the hot end of
>>>> the resonance pot in the feedback path. Some of the Roland Juno filters
>>>> implement this technique.
>>
>>
>>>>
>>
>>>> Another technique is to take some of the filter output from the hot end
>>>> of the resonance pot, plus an additional amount from the wiper of the
>>>> resonance pot. See tb-303 schematic for an example of this gain makeup
>>>> method.
>>
>>
>>>>
>>
>>>> Compensating for the potentially deafening squeal when the filter goes
>>>> into self-oscillation is different. The amplitude of sustained
>>>> self-oscillation is determined by the behaviour of non-linearities in the
>>>> filter. Eg, hard or soft clipping. If the self-oscillation level is much
>>>> louder than the oscillator signal you can try driving the filter with a
>>>> hotter signal, and/or introducing soft clipping into the filters feedback
>>>> path. Some of Rolands SH synths use back to back diodes in the feedback path
>>>> around the filter to limit the amplitude of self-oscillation.
>>
>>
>>>>
>>
>>>> You didn't say what model your synth is, or the filter topology. So I
>>>> hope this info is some help to you.
>>
>>
>>>>
>>
>>>> -Richie,
>>
>>
>>>>
>>
>>>> Sent from my Sony Ericsson Xperia ray
>>
>>
>>>>
>>
>>>> Nicholas Keller <niroke at tampabay.rr.com> wrote:
>>
>>
>>>>
>>
>>>>> I have a synth that seriously jumps in volume when its filter
>>>>> oscillates.  I know that later generation chips had a way of compensating
>>>>> for this change, keeping the level constant regardless of the resonance
>>>>> setting.  Is there a circuit I could add to my synth that would perform this
>>>>> compensation?  I could use an audio limiter perhaps, but I'm concerned about
>>>>> clipping and squaring off the waveform.  I'm wondering if anyone knows how
>>>>> these chips did it, like the SSM2044 and the later CEMs, and if I could
>>>>> build a similar add-on circuit for my synth (and possibly other modules I
>>>>> have, like the Doepfer A-102).
>>
>>
>>>>>
>>
>>>>> I have an idea of adding an envelope follower (inverted) and VCA.
>>>>> Maybe that's how it was originally done.
>>
>>
>>>>>
>>
>>>>> Thanks,
>>
>>
>>>>> Nick
>>
>>
>>>>>
>>
>>>>>
>>
>>>>> _______________________________________________
>>
>>
>>>>> Synth-diy mailing list
>>
>>
>>>>> Synth-diy at dropmix.xs4all.nl
>>
>>
>>>>> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy [6]
>>
>>
>>>>
>>
>>>> _______________________________________________
>>
>>
>>>> Synth-diy mailing list
>>
>>
>>>> Synth-diy at dropmix.xs4all.nl
>>
>>
>>>> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy [6]
>>
>>
>>>
>>
>>> _______________________________________________
>>
>>
>>> Synth-diy mailing list
>>
>>
>>> Synth-diy at dropmix.xs4all.nl
>>
>>
>>> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy [6]
>>
>>
>> Links:
>> ------
>> [1]
>> http://niroke.blogspot.com/2013/03/octave-cat-srm-factory-modifications.html
>> [2] http://i.imgur.com/gtyGktQ.png
>> [3] http://www.youtube.com/watch?v=p_WkSj14C-k
>> [4] http://i.imgur.com/R028I6l.jpg
>> [5] http://manuals.fdiskc.com/flat/Roland%20SH101%20Service%20Manual.pdf
>> [6] http://dropmix.xs4all.nl/mailman/listinfo/synth-diy
>
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at dropmix.xs4all.nl
> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy